BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//wp-events-plugin.com//6.3//EN
TZID:Asia/Jerusalem
X-WR-TIMEZONE:Asia/Jerusalem
BEGIN:VEVENT
UID:29@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20200209T143000
DTEND;TZID=Asia/Jerusalem:20200209T153000
DTSTAMP:20201124T121630Z
URL:https://dds.technion.ac.il/iemevents/prediction-of-human-individual-be
havior/
SUMMARY:Prediction of human individual behavior [ \n Graduate Student S
eminar\n Seminars\n \n ]
DESCRIPTION:By: MSc Meghan Mergui\n Advisors: Tamir Hazan and Ori Plonsky\n
Where: Bloomfield 527 From:\nTechnion\nAbstract:\n\nPrediction of individ
ual human behavior is a fundamental task in many domains. For example\, fi
rms may want to offer individually tailored promotions to consumers. Our r
esearch aims to give better predictions of individual choice behavior\, fo
cusing on predicting choice behavior in simple abstract economic games. A
recent prediction tournament (CPC18) challenged researchers to predict beh
avior in such games\, in two competition tracks. In the first track\, the
challenge was to predict the average aggregate behavior of people in a new
game\, and in the second track\, the challenge was to predict the behavio
r of individual decision makers in familiar games. The results of CPC18 sh
owed impressive predictive performance in the first track\, when predictin
g the average agent’s behavior\, specifically using a combination of mac
hine learning and a psychological model. However\, when predicting individ
uals’ choice behavior\, the best prediction model was a naïve baseline
that predicts that each individual behaves the same as the “average agen
t” in the training set. Using deep learning models\, and by leveraging t
he success of the models designed for prediction of the average agent\, we
try to make an advancement in this task. Can we extract knowledge about i
ndividual behavior that goes beyond what is known about the population? Ca
n we learn psychological parameters of individuals from previous behavior
of those individuals? Can we use learnt psychological parameters to improv
e prediction on the individuals' future choice behavior?
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:6@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20200818T110000
DTEND;TZID=Asia/Jerusalem:20200818T120000
DTSTAMP:20200817T112458Z
URL:https://dds.technion.ac.il/iemevents/strategic-information-retrieval-t
he-mediator-and-agent-perspectives/
SUMMARY:Strategic information retrieval - the mediator and agent perspectiv
es [ \n Graduate Student Seminar\n \n ]
DESCRIPTION:By: PhD Gregory Goren\n Advisors: Prof. Oren Kurland and Prof.
Moshe Tennenholtz\n Where: https://technion.zoom. us/j/3800541616 From:\n\
nIn the Web retrieval setting\, many document authors are "ranking-incenti
vized". That is\, they are interested in having their documents highly ran
ked for some queries by search engines. To this end\, they often respond t
o rankings by introducing modifications to their documents (a.k.a.\, searc
h engine optimization). Hence\, the retrieval setting is competitive.\nA s
earch engine can be considered as a mediator. It connects users with infor
mation needs (which are represented via queries) and document authors (age
nts) whose pages might satisfy the information needs of the users. This bo
dy of work tackles both the mediator perspective and the agent perspective
in the competitive retrieval setting. We show an extensive analysis of th
e robustness of ranking functions to adversarial document manipulations. I
n addition\, we devise an automatic model for ranking-incentivized\, quali
ty preserving document manipulations. Finally\, we present an empirical an
alysis of the possibility of an herding effect in a competitive search set
ting. That is to say\, the agents are choosing similar strategies without
an explicit centralized direction.
CATEGORIES:Graduate Student Seminar
END:VEVENT
BEGIN:VEVENT
UID:15@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20200818T120000
DTEND;TZID=Asia/Jerusalem:20200818T130000
DTSTAMP:20201109T060732Z
URL:https://dds.technion.ac.il/iemevents/cluster-based-document-retrieval-
with-multiple-queries/
SUMMARY:Cluster-Based Document Retrieval with Multiple Queries [ \n Gra
duate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: MSc Kfir Bernstein\n Advisors: Oren Kurland\n Where: Zoom F
rom:\nTechnion\nThe merits of using multiple queries representing the same
information need to improve retrieval effectiveness have recently been
demonstrated in several studies. In this paper we present the first stud
y of utilizing multiple queries in cluster-based document retrieval\; tha
t is\, using information induced from clusters of similar documents to ra
nk documents. Specifically\, we propose a conceptual framework of retriev
al templates that can adapt cluster-based document retrieval methods\, or
iginally devised for a single query\, to leverage multiple queries. The ad
aptations operate at the query\, document list and similarity-estimate le
vels. Retrieval methods are instantiated from the templates by selecting\,
for example\, the clustering algorithm and the cluster-based retrieval m
ethod. Empirical evaluation attests to the merits of the retrieval templa
tes with respect to very strong baselines: state-of-the-art cluster-based
retrieval with a single query and highly effective fusion of document li
sts retrieved for multiple queries. In addition\, we present findings abo
ut the impact of the effectiveness of queries used to represent an inform
ation need on (i) cluster hypothesis test results\, (ii) percentage of re
levant documents in clusters of similar documents\, and (iii) effectivene
ss of state-of-the-art cluster-based retrieval methods.\n\nJoint work wit
h Dr. Fiana Raiber (Yahoo Research) and Prof. J. Shane Culpepper (RMIT)
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:60@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20200915T110000
DTEND;TZID=Asia/Jerusalem:20200915T120000
DTSTAMP:20210224T094649Z
URL:https://dds.technion.ac.il/iemevents/online-distributionally-robust-op
timization/
SUMMARY:Online Distributionally Robust Optimization [ \n Graduate Stude
nt Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Ido Yerenburg\n Advisors: Dr. Shimrit Shtern\n Where:
ZOOM From:\nTechnion\nAbstract:\n\nWe are interested in exploring stocha
stic optimization problems where the distribution of the uncertainty is un
known but random data from the distibution is continuously becoming avaial
ble. When the entire data is available in advance it is known that Sample
Average Approximation (SAA) tends to overfit the data\, and recently data-
driven distributionally robust optimization (DRO) approaches have been sug
gested as an alternative. We focus on one such DRO with ambiguity set base
d on the Wasserstein metric (WDRO)\, which is known to provide finite-samp
le guarantees but becomes more computationally demanding the more data poi
nts exists. Thus\, we explore extending the use of online methods\, such a
s online gradient descent and online mirror descent\, used to approximate
SAA to the DRO setting under assumptions of both bounded and unbounded but
light tailed distributions. We show that this online version of the WDRO
converges to the true stochatic problem almost surely and maintains some o
f WDRO probabilistic guarantees. In addition\, we show numeric results dem
onstrating the benefits and disadvantages of using online-DRO compared to
online-SAA and the offline versions of both algorithms.\n\nZOOM Link\n\nht
tps://technion.zoom.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:9@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20201101T143000
DTEND;TZID=Asia/Jerusalem:20201101T153000
DTSTAMP:20201109T060808Z
URL:https://dds.technion.ac.il/iemevents/competition-among-contests-a-safe
ty-level-analysis/
SUMMARY:“Competition Among Contests: a Safety Level Analysis" [ \n Gr
aduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: Omer Shiran-Shvarzbard\n Advisors: Ron Lavi\n Where: Zoom
From:\nTechnion\nAbstract:\nWe study a competition among two contests\, w
here each contest designer aims to attract as much effort as possible. Suc
h a competition exists in reality\, e.g.\, in crowd-sourcing websites. Our
results are phrased in terms of the ``relative prize power'' of a contest
\, which is the ratio of the total prize offered by this contest designer
relative to the sum of total prizes of the two contests. When contestants
have a quasi-linear utility function that captures both a risk-aversion ef
fect and a cost of effort\, we show that a simple contest attracts a total
effort which approaches the relative prize power of the contest designer
assuming a large number of contestants. This holds regardless of the conte
st policy of the opponent\, hence providing a ``safety level'' which is a
robust notion similar in spirit to the max-min solution concept.\n\nSemina
r Zoom Link
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:14@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20201108T143000
DTEND;TZID=Asia/Jerusalem:20201108T153000
DTSTAMP:20201109T060912Z
URL:https://dds.technion.ac.il/iemevents/waiting-time-prediction-in-queuei
ng-systems-via-learning-models/
SUMMARY:Waiting Time Prediction in Queueing Systems via Learning Models [ \
n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc Elisheva Chocron\n Advisors: Paul Feigin\, Izack Cohen
\n Where: Zoom From:\nTechnion\nAbstract\n\nWe research approaches for pr
edicting waiting times of customers within service systems. Our focus is o
n call centers in which accurate waiting time predictions may enable impro
ved work-force management and lead to increased customer satisfaction.\n\n
Standard approaches for design and management of service systems that incl
ude customers\, servers and queues rely on Queueing Theory (QT). In partic
ular\, QT is often used for predicting waiting times which is the focus of
this work.\n\nQT is often criticized since its formulas are based on fund
amental assumptions regarding the model describing a given system — assu
mptions which are usually violated.\n\nTo overcome this issue\, we hypothe
size that additional external characteristics (“features”) may be requ
ired in order to provide an accurate prediction of the waiting time for ma
ny real-life systems. We further hypothesize that a Machine Learning (ML)
model incorporating those characteristics will improve the performance of
waiting time predictions. Consequently\, in this research we explore the u
se of ML-based algorithms for waiting time prediction and investigate to w
hat extent they overcome the limits of QT methods which are typically base
d on “simple model” assumptions. In a series of experiments\, we use m
ultiple ML- and QT-based models to predict the waiting times of customers
within simple (“synthetic”) and real-life call centers.\n\nWe analyze
the performance of the different models\, in order to determine for which
methods and under what conditions ML performance compares favorably with Q
T performance.\n\nSeminar Zoom Link
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:22@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20201115T143000
DTEND;TZID=Asia/Jerusalem:20201115T153000
DTSTAMP:20201109T061019Z
URL:https://dds.technion.ac.il/iemevents/self-stabilizing-algorithms-in-th
e-stone-age-model/
SUMMARY:Self-Stabilizing Algorithms in the Stone Age Model [ \n Graduat
e Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: Eyal Keren\n Advisors: Yuval Emek\n Where: Zoom From:\nTe
chnion\nAbstract:\n\nIntroduced by Emek and Wattenhofer (PODC 2013)\, the
\\emph{stone age (SA)} model suggests an abstraction for network algorithm
s distributed over randomized finite state machines.\n\nRecent works demon
strate that the weak computational environment provided by the SA model is
sufficient for efficient solutions to some core problems in distributed c
omputing\, but they do so under the (somewhat unrealistic) assumption of f
ault free computations.\nIn this talk\, we initiate the study of \\emph{s
elf-stabilizing} SA algorithms that are guaranteed to recover from any com
bination of transient faults.\nSpecifically\, we develop efficient self-st
abilizing SA algorithms for the \\emph{leader election} and \\emph{maximal
independent set} problems in bounded diameter graphs.\nThese algorithms r
ely on a novel self-stabilizing \\emph{synchronizer} for SA algorithms ope
rating in such graphs.\n\nSeminar Zoom Link
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:33@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210110T143000
DTEND;TZID=Asia/Jerusalem:20210110T153000
DTSTAMP:20201230T091933Z
URL:https://dds.technion.ac.il/iemevents/eptas-for-load-balancing-problem-
on-parallel-machines-with-a-non-renewable-resource/
SUMMARY:EPTAS for Load Balancing Problem on Parallel Machines with a Non-re
newable Resource [ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Jaykrishnan G.\n Advisors: Prof. Asaf Levin \n Wher
e: Zoom From:\nTechnion\nAbstract\n\nThe problem considered is the non-pr
eemeptive scheduling of ‘n’ independent jobs that consume a resource (
which is non-renewable and replenished regularly) on ‘m’ parallel mach
ines. The input defines the speed of machines\, size of jobs\, the quantit
y of resource required by the jobs\, and the replenished quantities and re
plenishment dates of the resource. Every job can be processed only after t
he required quantity of the resource is allocated to the job. The objectiv
e function is the minimization of the convex combination of the makespan a
nd an objective that is equivalent to the lp-norm of the vector of loads o
f the machines.\n\nA polynomial time approximation scheme (PTAS) for a giv
en problem is a family of approximation algorithms such that the family ha
s a (1+ε)-approximation algorithm for any ε>\;0. An efficient polynomi
al time approximation scheme (EPTAS) is a PTAS whose time complexity is up
per bounded by the form f(1/ε) ⋅ poly(n) where ‘f’ is some computab
le (not necessarily polynomial) function and poly(n) is a polynomial of th
e length of the (binary) encoding of the input. We establish the existence
of an EPTAS for the problem. The EPTAS will first apply rounding steps to
structure the input. We characterize structural properties of near optima
l solutions and use it to formulate a Mixed-integer linear program (MILP)
with a constant number of integer variables. Based on the optimal solution
of this MILP\, a feasible solution for the scheduling problem will be com
puted and we show that this solution approximates the optimal solution by
a factor of (1+ ε).\n\nZoom link
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:47@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210117T143000
DTEND;TZID=Asia/Jerusalem:20210117T153000
DTSTAMP:20210110T094635Z
URL:https://dds.technion.ac.il/iemevents/queueing-inference-for-process-pe
rformance-analysis-with-missing-life-cycle-data/
SUMMARY:Queueing Inference for Process Performance Analysis with Missing Li
fe-Cycle Data [ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Guy Berkenstadt\n Advisors: Prof. Avigdor Gal\n Where
: Zoom link From:\nTechnion\nAbstract:\n\nMeasuring key performance indica
tors\, such as queue lengths and waiting times\, using event logs serve fo
r improvement of resource-driven business processes. However\, existing te
chniques assume the availability of complete life cycle information\, incl
uding the time a case was scheduled for execution (aka arrival times). Yet
\, in practice\, such information may be missing for a large portion of th
e recorded cases.\n\nIn this talk\, I will propose a methodology to addres
s missing life-cycle data by incorporating predicted information in busine
ss processes performance analysis. The approach builds upon techniques fro
m queueing theory and leverages supervised learning to accurately predict
performance indicators based on an event log with missing data. I will als
o present experimental results to demonstrate the effectiveness of the app
roach on synthetic and real-world data.\n\nZoom link\n\n \;\n\nhttps:/
/technion.zoom.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:45@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210124T143000
DTEND;TZID=Asia/Jerusalem:20210124T153000
DTSTAMP:20210105T124457Z
URL:https://dds.technion.ac.il/iemevents/married-women-labor-supply-and-di
fferences-between-ethnic-groups-in-israel/
SUMMARY:Married Women Labor Supply\, and Differences Between Ethnic Groups
in Israel. [ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Sabaa Jabali-Serhan \n Advisors: Doctor Jacob Schwart
z and Professor (emeritus) Benjamin Bental\n Where: Zoom From:\nTechnion\n
Abstract:\n\nThis study proposes empirically examines what affects married
women labor supply in Israel\, focusing on the differences between the be
havior of Arab and Jewish married women. The major contribution of this re
search is that unlike most other studies\, which examine the participation
decision\, i.e. whether to enter the labor market or not\, I examine the
effect of individual and familial factors on hours supplied over the years
. Using the PUF version of the cross-sectional Israeli Income Surveys for
the 1997 to 2011 time period\, the results show that the expected monthly
hours supplied of married Arab and Jewish women are positively and signifi
cantly affected by additional education years\, are concave with age\, are
negatively and significantly affected by children aged 0-14\, and negligi
bly affected by other sources of income. These findings are similar to the
existing in literature. However\, there are major differences in the esti
mated coefficients for Arab and Jewish married women. For Arab women\, the
re is high volatility in the estimated explanatory variables over the vari
ous surveys. In addition\, the study adopts a pseudo panel approach to est
imate the elasticities of monthly hours of work with respect of wages\, cr
oss wages and income. The estimated wage elasticity for Arab and Jewish ma
rried women during the sample years is almost identical. One percent chang
e in the wage raises Arab married women expected monthly hours by 0.59% an
d that of Jewish married women by 0.60%. Furthermore\, the estimated elast
icities are larger for better educated women both for Arab and Jewish marr
ied women.\n\nZoom\nhttps://technion.zoom.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:53@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210124T160000
DTEND;TZID=Asia/Jerusalem:20210124T170000
DTSTAMP:20210114T185342Z
URL:https://dds.technion.ac.il/iemevents/the-effect-of-national-security-e
vents-on-life-satisfaction-of-different-section-of-the-israeli-society/
SUMMARY:The effect of national security events on life satisfaction of diff
erent section of the Israeli society [ \n Graduate Student Seminar\n
Seminars\n \n ]
DESCRIPTION:By: M.Sc Dana Masharka\n Advisors: Prof. Bar-Ilan Avner\, Dr. A
iche Avishay\n Where: ZOOM From:\nTechnion\nAbstract:\n\nThe goal of this
research is to study factors affecting life satisfaction and financial sat
isfaction of Israeli citizens. In particular\, the 2006 Lebanon war and th
e 2014 Israel-Gaza conflict and their impact on Jews and Arabs. To do this
\, Israel's annual Social Survey of the Central Bureau of Statistics for t
he years 2005-2008 and 2012-2016 has been analyzed.\nThe main statistical
methods used are the Ordered Probit model\, quadratic discriminant analysi
s (QDA) model\, and Tree-based model. In most previous studies of happines
s\, the methods used are Logit\, Linear regression\, and Ordered Probit. T
he Ordered Probit used here to estimate the effect of various variables on
life and financial satisfaction. An innovative contribution of this thesi
s is to test the prediction power of these variables by QDA and Tree-Based
models.\nSome results are similar to those found in previous research. We
find that financial/apartment/family relationship satisfaction\, health\,
and optimism have a great positive effect on life satisfaction. Lonelines
s hurts life satisfaction\, and family size has a positive effect on both
life and financial satisfaction. It seems that orthodox Jews are happier t
han any other religious intensity. Age has a U-shaped effect both on life
and financial satisfaction. For financial satisfaction\, work and income s
atisfaction have a significant positive effect\, and income increases fina
ncial satisfaction. Job security has a positive effect on financial satisf
action too. There is a difference in satisfaction’ especially financiall
y\, for individuals who vary in religiosity. Arabs are more financially sa
tisfied than Jews\, and Orthodox are more satisfied with financial and lif
e than any other religious intensity.\nWe find that geographic area and se
rving in IDF affected satisfaction in the 2006 Lebanon war\, but not in 20
14. For both security events\, the financial satisfaction affecting life s
atisfaction decreased in the year of the war. For the 2014 Israel-Gaza con
flict\, the variables feeling safe\, Arab\, and loneliness have the bigges
t effect. For financial satisfaction\, the income satisfaction is the lowe
st in the years of the military conflicts\, and interestingly the variable
Arabs has the smallest effect on financial satisfaction in the years of t
he security events.\n\nZoom Link\nhttps://technion.zoom.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:50@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210127T150000
DTEND;TZID=Asia/Jerusalem:20210127T160000
DTSTAMP:20210112T190118Z
URL:https://dds.technion.ac.il/iemevents/shared-micro-depot-network/
SUMMARY:Shared micro-depot network [ \n Graduate Student Seminar\n
Seminars\n \n ]
DESCRIPTION:By: MSc Leonardo N. Rosenberg\n Advisors: Prof. Yale T. Herer\n
Where: ZOOM From:\n\nAbstract:\n\nLast-mile logistics is both a source an
d cause of problems in urban areas\, especially problems related to traffi
c congestion\, unsustainable delivery modes\, and limited parking availabi
lity. In this context\, multiple sustainable logistics solutions have been
proposed. One of them is micro-depots (MDs)\, which can function as a con
solidation center and a collection-and-delivery point. During the EIT Urba
n Mobility-funded project S.M.U.D. (Shared Micro-depots for Urban pickup a
nd Delivery)\, the concept of a shared MD network with parcel lockers was
developed. Such networks enable multiple logistics service providers (LSPs
) and/or business partners to use an MD while minimizing their individual
costs and optimizing the use of urban space. We present case studies of su
ch shared MD networks operating in the cities of Helsinki (FI) and Helmond
(NL). We provide a framework for auxiliary businesses that can exploit th
e existing MD structure to offer services to the surrounding population. T
he case studies highlight the complexity of implementing such a solution\;
it requires stakeholders’ involvement and collaboration. We modeled the
distance traveled using the shared MD network with different distribution
policies\, with or without an urban consolidation center located in the s
uburban area\, and compared the results of the cities involved in the proj
ect with the traditional modus operandi. Our results show that cargo bikes
can perform most of the distance traveled in the system\, reaching up to
80% of the total. The implementation of this network also can reduce subur
ban distance traveled by 60%.\n\n \;\n\nZoom Link: https://technion.zo
om.us/j/3800541616\n\n \;\n\n \;
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:56@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210210T143000
DTEND;TZID=Asia/Jerusalem:20210210T153000
DTSTAMP:20210131T152935Z
URL:https://dds.technion.ac.il/iemevents/competition-affirmative-action-an
d-sabotage-negative-characteristics-in-contest-design-observed-in-horse-ra
cing-in-the-united-kingdom-in-2019/
SUMMARY:Competition: Affirmative Action and Sabotage. Negative Characteris
tics in Contest Design Observed in Horse Racing in the United Kingdom in 2
019 [ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: MS.C Ilan Rosenberg\n Advisors: Prof. Todd Kaplan\n Where:
ZOOM From:\nTechnion\nAbstract:\nAffirmative action is a policy designed t
o balance opportunities and create open conditions for competition. Althou
gh such policies usually seek to support certain demographic groups\, comm
only referred to as "minorities" or "weakened groups"\, they are also appl
ied in sports\, political campaigns\, rent seeking contests and more. Whil
e the goal of affirmative action is to improve and equalize opportunities\
, the literature shows that not having such policies leaves too much asymm
etry among players. This asymmetry causes incentive problems\, resulting i
n reduced levels of the general effort due to the "despair effect"\, where
weaker players have low expectations and not willing to invest effort and
stronger players feel no need to invest effort.\nIn such cases\, a contes
t designer might consider implementing an affirmative action policy\, whet
her by weakening the strong players (handicapping) or by strengthening the
weaker players (head start). This work examines whether affirmative actio
n\, designed to reduce the gaps in the competitive world\, might by trying
to encourage effort might actually\, motivate contestants to sabotage oth
er contestants.\nUsing a natural experiment\, with data from horse racing
in the United Kingdom in 2019\, I demonstrate how affirmative actions that
handicap favorite horses\, results in a more balanced playing field by gi
ving weaker horses higher winning probabilities. I also demonstrate that c
ases of sabotage and negative behavior between riders\, are more prevalent
in such races. Sabotage by the leading jockeys\, improves their position
by an average of 0.98 placings. I also show that stronger riders (the top
5% of UK jockeys)\, are in general 4.4 times more involved in cases of int
erference between riders than regular jockeys.\n\n \;\n\nZoom Link\n\n
https://technion.zoom.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:58@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210214T153000
DTEND;TZID=Asia/Jerusalem:20210214T163000
DTSTAMP:20210207T102858Z
URL:https://dds.technion.ac.il/iemevents/a-steiner-tree-perspective-on-top
ical-crawling/
SUMMARY:A Steiner Tree Perspective on Topical Crawling [ \n Graduate St
udent Seminar\n Seminars\n \n ]
DESCRIPTION:By: MSc Avihay Levi\n Advisors: Prof. Carmel Domshlak\n Where:
ZOOM From:\nTechnion\nAbstract:\n\nTopical\, also known as vertical\, sea
rch engines specialize in retrieval of information restricted to a certain
subject of information needs. The essential component of such a topical s
earch engine is a topical crawler\, a software agent that navigates the We
b in search for documents fitting the scope of its search engine. The comm
on approach to topical crawling appears to be best-first search over the d
igraph forming the Web\, prioritizing the links in its search queue by the
lexical similarity between the representation of the document embedding t
he link and the representation of the topic. In our work\, we re-examine t
his approach to topical crawling. First\, we show that the optimal footpri
nt of the topical crawling process constitutes a minimum directed Steiner
tree of the Web digraph\, with the on-topic documents being Steiner termin
als. Second\, having this "Steiner optimality" in mind\, we formalize a no
vel best-first approach that prioritizes search directions based on a Baye
sian inference model that continuously updates its estimates given evidenc
e collected during crawling. Our empirical comparative evaluation on real-
world large snapshots of the Web shows that the proposed approach substant
ially outperforms the standard technique for best-first topical crawling.\
n\nZoom Link\n\nhttps://technion.zoom.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:59@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210224T143000
DTEND;TZID=Asia/Jerusalem:20210224T153000
DTSTAMP:20210221T123120Z
URL:https://dds.technion.ac.il/iemevents/expansion-and-generalization-of-t
he-literature-on-the-risk-averse-newsvendor-problem-by-integrating-finance
-results-with-the-risk-averse-newsvendor-problem/
SUMMARY:Expansion and generalization of the literature on the risk averse n
ewsvendor problem by integrating finance results with the risk-averse news
vendor problem [ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc Gad Gal (Gadi) Ezer\n Advisors: Professor Yale T. Here
r \n Where: ZOOM From:\nTechnion\nAbstract:\nGlobalization and e-commerce
have led to increased levels of competition and steadily shorter product
life cycles. The newsvendor (NV) model is classically used for inventory m
anagement of short life-cycle product. It contends with demand uncertainty
that leads to uncertain performances\, as a result risk emerges under whi
ch decision are made. While decision makers tend to avoid risk for high pr
ofit products (e.g.\, new products)\, the traditional solution of the NV p
roblem ignores the underlying risk. Mean-variance (MV) analysis is a funda
mental theory of risk management in finance.MV analysis is notable for bei
ng implementable and providing good recommendations even without knowing t
he utility function. Inspired by the Modern Portfolio Theory (MPT) we inve
stigate the set of efficient solutions (i.e.\, efficient frontier) of the
NV problem\, using the MV analysis methodology with a modification to avoi
d its pitfall.\nWe focus on the special case of uniformly distributed dema
nd where identical locations are independent such that their decisions do
not affect their demands. While MPT uses diversification to create an effi
cient frontier\, we show that the efficient frontier of the NV problem exi
sts also for a single location. Defining a combination of multiple locatio
ns’ strategies as a portfolio\, we show that the efficient frontier of t
he multilocation NV problem\, includes only efficient strategies of the in
dividual problems. Moreover\, when the efficient frontier of the individua
l problems is convex\, a portfolio is efficient if and only if it includes
identical efficient strategies of the individual problems.\nMature produc
ts’ demands are often stabilized such that their profit fluctuation is n
egligible and so we consider location’s choice to solely sell it\, as a
risk-free strategy. New products have a higher expected profit\, but conta
in uncertainty in their demand. MPT showed that the set of all combination
s of fractions between a risk-free strategy and a specific efficient portf
olio (i.e.\, the market portfolio)\, results in superior efficient frontie
r. Capital Asset Pricing Model (CAPM) set a criterion to make decision abo
ut adding products to an existing efficient portfolio. While both MPT and
CAPM focus is adding products to an existing efficient portfolio\, we inve
stigate the effect of replacing locations’ strategies with risk-free str
ategies—within an existing portfolio. We demonstrate that we can gain la
rge reduction in risk by sacrificing only a small amount of the expected p
rofit.\n\nZOOM LINK\n\nhttps://technion.zoom.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:61@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210404T143000
DTEND;TZID=Asia/Jerusalem:20210404T153000
DTSTAMP:20210304T130214Z
URL:https://dds.technion.ac.il/iemevents/opm-model-based-iot-and-internet-
of-robotic-things/
SUMMARY:OPM Model-Based IoT and Internet of Robotic Things [ \n Graduat
e Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: PhD Hanan Kohen\n Advisors: Prof. Dov Dori\n Where: Zoom Fr
om:\nTechnion\nThe realization that models can and should become the centr
al artifact in engineered systems’ lifecycles has been gaining momentum
in recent years\, giving rise to model-based systems engineering (MBSE) as
an evolving SE field. Conceptual modeling at the system level precedes ma
thematical\, physical\, geometrical\, and detailed disciplinary modeling a
nd design. Object-Process Methodology (OPM) ISO 19450 is a conceptual and
quantitative executable MBSE modeling language and methodology.\nThe Inter
net of Things (IoT) capitalizes on the Internet to ubiquitously enable inf
ormation exchange among daily physical objects that are integrated into co
mplex interconnected networks\, providing for improved resource utilizatio
n and other benefits to humanity. Within the conceptual framework of IoT\,
the Internet of Robotic Things (IoRT) concerns the integration of humans
and autonomous agents. i.e.\, robots. The main function of IoRT applicatio
ns is monitoring robot states and processes and coordinating multiple robo
ts. As the IoT and IoRT continue to penetrate into homes and businesses\,
and into national and global infrastructures\, there is a growing need for
effective development methodologies and tools that account for performanc
e\, security\, safety\, cost and the environment.\nIn this research\, we h
arness OPM and its modeling software environment OPCloud to develop IoT an
d IoRT systems from the very early stages of concept development through a
rchitecting\, all the way to detailed design\, optimization\, simulating\,
deployment and prototyping. Using an MBSE approach\, aspects of interest\
, such as security or cost can be modeled and analyzed explicitly\, enabli
ng trade-space exploration and Pareto front optimization with respect to t
hese aspects.\n\n \;\n\nZoom Link\n\nhttps://technion.zoom.us/j/380054
1616
ATTACH;FMTTYPE=image/jpeg:https://dds.technion.ac.il/wp-content/uploads/20
19/12/Hanan-2.png
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:62@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210411T143000
DTEND;TZID=Asia/Jerusalem:20210411T153000
DTSTAMP:20210404T103344Z
URL:https://dds.technion.ac.il/iemevents/analysis-of-ecommerce-best-seller
-lists/
SUMMARY:Analysis of eCommerce best seller lists [ \n Graduate Student S
eminar\n Seminars\n \n ]
DESCRIPTION:By: MSc Lital Kuchy\n Advisors: Prof. Oren Kurland\n Where: ZO
OM From:\nTechnion\nAbstract: \nOnline shopping is one of the fastest gr
owing markets of the 21st century. Today we can buy almost everything onli
ne\, from groceries to house furniture. \nAs the market grows\, more and
more manufacturers offer their goods online. If one is looking to buy a sm
art watch for example\, a simple search on Amazon.com alone\, \none of
the largest e-commerce sites in the world\, would result in over 3000 diff
erent options. Such information overload makes it impossible for the custo
mer to manually evaluate all options. \nTo help the customer make a purch
ase decision\, many eCommerce sites offer different product ranking option
s.\nIn this work\, we analyze different aspects of best seller ranked list
s.\nJoint work with Dr. David Carmel\n \;\n\nZoom Link\nhttps://techni
on.zoom.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:65@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210418T143000
DTEND;TZID=Asia/Jerusalem:20210418T153000
DTSTAMP:20210407T082539Z
URL:https://dds.technion.ac.il/iemevents/studying-ranking-incentivized-web
-dynamics/
SUMMARY:Studying Ranking-Incentivized Web Dynamics [ \n Graduate Studen
t Seminar\n Seminars\n \n ]
DESCRIPTION:By: MSc Ziv Vasilisky\n Advisors: Prof. Oren Kurland\, Prof. Mo
she Tennenholtz \n Where: ZOOM From:\nTechnion\nAbstract:\nThe ranking inc
entives of many authors of Web pages play an important role in the Web dyn
amics. That is\, authors who opt to have their pages highly ranked for que
ries of interest often respond to rankings for these queries by manipulati
ng their pages\; the goal is to improve the pages' future rankings. Variou
s theoretical aspects of this dynamics have recently been studied using ga
me theory. However\, empirical analysis of the dynamics is highly constrai
ned due to lack of publicly available datasets. We present an initial such
dataset that is based on TREC's ClueWeb09 dataset. Specifically\, we used
the WayBack Machine of the Internet Archive to build a document collectio
n that contains past snapshots of ClueWeb documents which are highly ranke
d by some initial search performed for ClueWeb queries. Temporal analysis
of document changes in this dataset reveals that findings recently present
ed for small-scale controlled ranking competitions between documents' auth
ors also hold for Web data. Specifically\, documents' authors tend to mimi
c the content of documents that were highly ranked in the past\, and this
practice can result in improved ranking.\n\nZoom Link\n\nhttps://technion.
zoom.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:66@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210425T143000
DTEND;TZID=Asia/Jerusalem:20210425T153000
DTSTAMP:20210412T081245Z
URL:https://dds.technion.ac.il/iemevents/behavioral-study-of-bidding-on-pe
er-reviews/
SUMMARY:Behavioral study of bidding on peer reviews [ \n Graduate Stude
nt Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Inbal Rozencweig \n Advisors: Reshef Meir\n Where: ZO
OM From:\nTechnion\nAbstract:\n\nPeer review is an essential step to ensur
e the quality of papers accepted to large conferences. In our work\, we wi
ll examine the different behaviors of reviewers in the bidding stage\, whe
re our main focuses will be on how reviewer's behavior is affected by pape
r's display order\, reviewers preference\, and 'points' associated with ea
ch paper that ostensibly reflects his demand\; the higher the 'points'\, t
he higher the chance to get the paper. Our empirical methodology consisted
of a bidding platform\, where reviewers bid on a set of papers from a lar
ge list of submitted papers.\n\nWe implemented a paper review bidding syst
em where we conducted a series of controlled lab experiments. In each expe
riment\, we isolated one variable and examined how it affects reviewers' b
ehavior in three key parameters - order bias\, price bias\, and private pr
eference bias. We found that presenting a score for each paper significant
ly changes the behavior of reviewers and incentivized them to choose paper
s with a higher score\, i.e. papers with a higher chance of getting. Also\
, we found that there is a bias in selecting papers that are displayed at
the top of the list\, and that reviewers choose papers that match their de
clared preferences. Finally\, we fitted a regression model for the results
of each experiment which presents the factors influencing the behavior of
the reviewers and enables prediction about future bids.\n\n \;\n\nZoo
m Link\n\nhttps://technion.zoom.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:69@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210502T143000
DTEND;TZID=Asia/Jerusalem:20210502T153000
DTSTAMP:20210418T064518Z
URL:https://dds.technion.ac.il/iemevents/bilevel-optimization-problems-met
hodology-and-first-order-methods-for-solving-convex-non-smooth-and-non-str
ongly-convex-problems/
SUMMARY:Bilevel optimization problems- methodology and first-order methods
for solving convex non-smooth and non-strongly convex problems. [ \n G
raduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Lior Doron\n Advisors: Dr. Shimrit Shtern\n Where: ZO
OM From:\nTechnion\nAbstract:\nSimple bilevel problems are optimization pr
oblems in which we want to find an optimal solution to an inner problem th
at minimizes an outer objective function. Such problems appear in many mac
hine learning and signal processing applications as a way to eliminate und
esirable solutions. However\, since these problems do not satisfy regulari
ty conditions\, they are often hard to solve exactly and are usually solve
d via regularization (e.g LASSO and ridge regression). In the past few yea
rs\, several algorithms were proposed to solve these bilevel problems dire
ctly and provide a rate for obtaining feasibility\, assuming that the oute
r function is strongly convex. In our work\, we suggest a new approach tha
t is designed for bilevel problems with simple outer functions\, such as t
he l1 norm\, which are not required to be either smooth or strongly convex
. In our new Iterative Approximation and Level-set Expansion (ITALEX) appr
oach\, we alternate between expanding the level-set of the outer function
and approximately optimizing the inner function over this level- set. ITAL
EX guarantees that at each iteration of the algorithm the outer objective
function is super-optimal\, a property that is not known for other bilevel
algorithms. We show that optimizing the inner function through first-orde
r schemes such as proximal gradient and generalized conditional gradient r
esults in a feasibility convergence rate of O(1/k)\, which is a rate only
shown to be obtained for a smooth strongly convex outer functions. Assumin
g more restrictive settings which still holds for important cases (like Ba
sis Pursuit)\, it results a convergence rate of O(1/k^2).\n\nZoom Link\n\n
https://technion.zoom.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:71@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210509T143000
DTEND;TZID=Asia/Jerusalem:20210509T153000
DTSTAMP:20210505T080148Z
URL:https://dds.technion.ac.il/iemevents/digital-screen-effect-on-behavior
-and-decision-making-insights-from-the-restaurant-industry/
SUMMARY:Digital Screen Effect on Behavior and Decision Making\, Insights fr
om the Restaurant Industry [ \n Graduate Student Seminar\n Seminar
s\n \n ]
DESCRIPTION:By: M.Sc. Gal Hirsh Ben-Mellech\n Advisors: Prof. Doron Kliger
and Dr. Benjamin Bachi \n Where: ZOOM From:\nTechnion\nAbstract:\nThe digi
tal transformation has gained prominence in recent years. The world as we
know it is changing\, and it has become substantial to develop an in-depth
understanding of the digitization effects on individuals' decision-making
and behavior. In our study\, we examine how the digital screen affects an
individual's decision-making while focusing on the restaurant industry. B
y conducting a field experiment that we divided into two sub-experiments\,
we defined several interventions\; Utilizing a digital-order cart and sec
tioning the dessert menu into categories. We demonstrate that customers te
nd to order more when they use a digital cart. Furthermore\, customers who
have used the digital cart tend to make fewer changes to their order. How
ever\, we were unable to show that the digital cart affects an individual'
s ordering standard dishes. In addition\, we were unable to infer a clear
conclusion and significance about the effect of sectioning the dessert men
u into categories on individual decision-making in terms of order size and
standard order. There is a place for further research in this interventio
n.\n\n \;\n\nWe completed our experiment before the outbreak of the CO
VID-19 pandemic\, and in light of the widespread use of digital tools in t
his period\, we suggest to continue studying this field extensively.\n\n&n
bsp\;\n\nZoom Link\nhttps://technion.zoom.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:72@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210523T143000
DTEND;TZID=Asia/Jerusalem:20210523T153000
DTSTAMP:20210509T065050Z
URL:https://dds.technion.ac.il/iemevents/walking-direction-estimation-usin
g-accelerometer-and-magnetic-sensors-a-deep-network-based-framework/
SUMMARY:Walking Direction Estimation using Accelerometer and Magnetic Senso
rs: a Deep Network-Based Framework [ \n Graduate Student Seminar\n
Seminars\n \n ]
DESCRIPTION:By: M.Sc. Adi Manos\n Advisors: Prof. Tamir Hazan and Prof. Itz
ik Klein\n Where: ZOOM From:\nTechnion\nAbstract:\n\nSmartphone-based iner
tial and magnetic sensors can be the basis for pedestrian navigation\, whe
never external positioning signals are limited or unavailable. Such naviga
tion solutions are typically accomplished by a practice known as pedestria
n dead reckoning\, wherein step length and heading angle are estimated to
form the horizontal trajectory of the user. One of the main challenges in
these methods is the unknown angular misalignment between walking directio
n and device orientation\, which imposes great difficulty in estimating th
e pedestrian's true heading.\nIn this work\, based on accelerometer and ma
gnetic sensors\, a new framework is established to estimate the user's hea
ding. It comprises a novel deep network architecture\, where temporal conv
olutions and multi-scale attention layers are trained to extract the walki
ng direction vector in the sensors' coordinate frame\, by using accelerati
on signals in a rotation-invariant manner. On top of that\, a unique geome
tric model is derived\, in which gravity and geomagnetic measurements are
combined with the estimated motion vector\, for calculating the pedestrian
's heading angle in the north and east coordinates. The proposed model is
trained and validated\, in a user-dependent approach\, through extensive e
xperiments of natural walking activity with commercial smartphones.\n\n&nb
sp\;\n\nZoom Link\n\nhttps://technion.zoom.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:78@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210606T133000
DTEND;TZID=Asia/Jerusalem:20210606T143000
DTSTAMP:20210601T202516Z
URL:https://dds.technion.ac.il/iemevents/incorporating-system-dynamics-int
o-opm-for-improved-model-analysis/
SUMMARY:Incorporating System Dynamics into OPM for Improved Model Analysis
[ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: MSc Keren-Or Rosenbaum\n Advisors: Prof. Dov Dori\n Where:
ZOOM From:\nTechnion\nAbstract:\n\n\n\nThe digital transformation that has
been shifting the world has urged policymakers to look for tools that mod
el this highly complex web of systems in a structured and easily understan
dable way to assist in designing better operating policies and guide effec
tive change. System Dynamics\, SysD (Forrester\, 1971)\, is a recognize
d method for modeling and predicting the dynamic behavior of complex syste
ms. In this research\, we devise a way to convert SysD models into mod
els in Object-Process Methodology (OPM\, ISO 19450)\, a conceptual and com
putational modeling language and model-based systems engineering method
ology. Using a real case study from the COVID-19 pandemic in Chile and oth
er SysD models of disease spreading and patient caring\, we have concluded
that a fully automated conversion is not possible\, because the SysD mode
ls contain implicit entities and assumptions that require the active invo
lvement of a human modeler to interpret the SysD model for creating an
elaborated OPM model. Yet\, we found a set of SysD constructs that can be
systematically converted into OPM constructs\, and these help modelers tra
nslate SysD models and evolve them into high-quality executable OPM models
that are both qualitative and quantitative.\n\n\nZoom Link\nhttps://techn
ion.zoom.us/j/3800541616\n\n\n
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:77@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210606T143000
DTEND;TZID=Asia/Jerusalem:20210606T153000
DTSTAMP:20210530T115816Z
URL:https://dds.technion.ac.il/iemevents/search-for-knowledge-grounded-res
ponses-in-retrieval-based-chatbots/
SUMMARY:Search for Knowledge-Grounded Responses in Retrieval-Based Chatbots
[ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: MSc Itay Harel\n Advisors: Prof. Oren Kurland\n Where: ZOOM
From:\nTechnion\nAbstract:\n\nBuilding a conversational system is conside
red a hardcore research problem. To have such a system is a research inter
est as well as commercial interest of E-commerce companies especially nowa
days\, in the big data era. Most of the advanced models\, mainly in the fi
eld of Deep Neural Networks\, focus on response generation but still suffe
r from generic responses. In this work\, we present novel methods for sear
ching knowledge-grounded responses. Empirical evaluation demonstrates the
effectiveness of these methods.\n\n \;\n\nZoom Link:\nhttps://technion
.zoom.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:82@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210613T143000
DTEND;TZID=Asia/Jerusalem:20210613T153000
DTSTAMP:20210611T032027Z
URL:https://dds.technion.ac.il/iemevents/symbols-in-support-of-depth-and-d
istance-estimation-implications-to-ar-displays/
SUMMARY:Symbols in support of depth and distance estimation: Implications t
o AR Displays [ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Carmel Zolkov\n Advisors: Prof. Avi Parush\n Where: Z
OOM From:\nTechnion\nAbstract:\n\nThere is a need of technologies that cou
ld support detection and identification of elements and objects in the env
ironment\, and facilitate spatial orientation in a complex and critical si
tuation and environments. Specifically\, when it comes to navigation and o
rientation\, there is a need to estimate depth and distances. In this rese
arch\, we focused on symbols that can be used in augmented reality display
s to support distance and depth estimation\, in order to establish spatial
orientation in different situations.\n\n \;\n\nThe research approach
was based on the development of realistic scenarios in a virtual reality u
rban environment and simulating AR displays\, in which participant perform
ed depth and distance estimation tasks in different spatial configurations
of buildings.\n\nThe main objectives of our research were:\n\n· Develop
and test symbols that best convey depth and distance in support of orienta
tion.\n\n· Define\, using psychophysical methods\, the behavior of symbol
s.\n\n· Test the effectiveness of the symbols in conveying distance in a
more cluttered urban environment\, by adding a symbol to the target\, whil
e the targets were static or dynamic\, and occluded or not.\n\nOur main re
sults indicate that people can adjust geometric parameters of a symbol (e.
g. length of a line or diameter of a circle) to express the distance they
estimated\, and that a linear function could well fit the relations betwee
n estimated distance and adjusted parameter of the symbol.\n\nWe also disc
overed that when estimating distances to elements in an environment presen
ted on a 2D display\, there is a consistent underestimation of distances\,
as well as increasing variation in the estimations\, as a function of the
real distance.\n\nWe tested the effectiveness of a progressive bar and a
circle as symbols\, and discovered that the use of the symbol’s psychoph
ysical functions does improve the accuracy of distance estimation. Specifi
cally\, accuracy of distance estimation improved with significant diminish
ing of the underestimation as a function of the distance\, that accuracy w
as better with non-occluded targets\, and specifically with a moving targe
t.\n\nThe findings are discussed in terms of their theoretical and practic
al implications.\n\nZoom Link\n\nhttps://technion.zoom.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:79@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210613T153000
DTEND;TZID=Asia/Jerusalem:20210613T163000
DTSTAMP:20210603T074851Z
URL:https://dds.technion.ac.il/iemevents/the-reactive-hidden-markov-model-
real-time-estimation-of-customer-satisfaction-in-contact-centers/
SUMMARY:The Reactive Hidden Markov Model: Real-Time Estimation of Customer
Satisfaction in Contact Centers [ \n Graduate Student Seminar\n Se
minars\n \n ]
DESCRIPTION:By: M.Sc. Lior Tony Landa \n Advisors: Prof. Galit Yom-Tov \n
Where: ZOOM From:\nTechnion\nAbstract:\n\nCustomer satisfaction is a key p
erformance indicator\, usually measured and analyzed using retrospective s
urveys. Retrospective analysis\, however\, offers limited value since the
organization cannot use it to address problems as they arise. Rather than
waiting for a conversation to end\, we are creating prediction models that
estimate customer satisfaction in real-time.\n\nWe utilize technical deve
lopments in sentiment analysis tools as well as previous observations show
ing that customer sentiment during a conversation correlates with retrospe
ctive customer satisfaction ratings. Considering that real-time customer s
atisfaction is unknown\, we develop unsupervised learning classification m
odels\, based on hidden Markov models (HMMs) and sentiment analysis\, to c
ategorize the customer state at time t into some arbitrary state space. Th
ese states are then mapped to customer satisfaction scores using retrospec
tive prediction models of customer satisfaction.\n\nWe study two types of
HMMs. In addition to the classical HMM proposed by Baum (1966)\, we develo
p a new reactive-HMM that takes into account the agent reactions to custom
er behavior. We find that the reactive-HMM is more accurate at predicting
customer satisfaction in retrospect and hence is recommended for real-time
prediction.\n\nThis is joint work with Antonio Castellanos\, Yair Goldber
g\, and Galit Yom-Tov.\n\nZoom Link\n\nhttps://technion.zoom.us/j/38005416
16
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:86@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210620T143000
DTEND;TZID=Asia/Jerusalem:20210620T153000
DTSTAMP:20210614T111922Z
URL:https://dds.technion.ac.il/iemevents/learning-to-estimate-search-progr
ess-using-sequences-of-states/
SUMMARY:Learning to Estimate Search Progress Using Sequences of States [ \n
Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc Matan Sudry \n Advisors: DR. Erez Karpas\n Where: ZOOM
From:\nTechnion\nAbstract:\nMany problems of interest can be solved using
heuristic search algorithms. When solving a heuristic search problem\, we
are often interested in estimating search progress\, that is\, how much l
onger until we have a solution. Previous work on search progress estimatio
n derived formulas based on some relevant features that can be observed fr
om the behavior of the search algorithm. In this paper\, rather than manua
lly deriving such formulas we leverage machine learning to automatically l
earn more accurate search progress predictors. We train a Long Short-Term
Memory (LSTM) network\, which takes as input sequences of states expanded
by the search algorithm\, and predicts how far along with the search we ar
e. Importantly\, our approach still treats the search algorithm as a black
box and does not look into the contents of search states. An empirical ev
aluation shows our technique outperforms previous search progress estimati
on techniques.\n\n \;\n\nZoom Link\n\nhttps://technion.zoom.us/j/38005
41616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:85@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210627T143000
DTEND;TZID=Asia/Jerusalem:20210627T153000
DTSTAMP:20210614T052241Z
URL:https://dds.technion.ac.il/iemevents/robust-learning-in-networks/
SUMMARY:Robust learning in networks [ \n Graduate Student Seminar\n
Seminars\n \n ]
DESCRIPTION:By: PhD Segev Shlomov\n Advisors: Prof. Yakov Babichenko\n Wher
e: ZOOM From:\nTechnion\nAbstract:\n\nWe introduce the class of virtually
additive non-Bayesian learning heuristics to aggregating beliefs in social
networks. A virtually additive heuristic is characterized by a single fun
ction that maps a belief to a real number that represents the virtual beli
ef. To aggregate beliefs\, an agent simply sums up all the virtual beliefs
of his neighbors to obtain his new virtual belief.\nThis class of heurist
ics determines whether robust learning\, by any naive heuristic\, is possi
ble. That is\, we show that in a canonical setting with a binary state and
conditionally i.i.d. signals whenever it is possible to naively learn the
state robustly it is also possible to do so with a virtually additive heu
ristic.\nWe also extend our results to achieve network-robust learning wit
h dynamics that are based on local weights adjustments that agents assigne
d to each other in the network. These adjustments are based on the famous
Sinkhorn-Knopp matrix scaling algorithm.\n\n \;\n\nZoom Link\n\n \
;\n\nhttps://technion.zoom.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:94@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210711T143000
DTEND;TZID=Asia/Jerusalem:20210711T153000
DTSTAMP:20210707T074420Z
URL:https://dds.technion.ac.il/iemevents/flexible-entity-resolution-for-mu
ltiple-intents/
SUMMARY:Flexible Entity Resolution for Multiple Intents [ \n Graduate S
tudent Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc Bar Genossar\n Advisors: Prof. Avigdor Gal\n Where: ZO
OM From:\nTechnion\nAbstract: Entity resolution (ER)\, a longstanding pro
blem of data cleaning and integration\, aims at identifying different data
records that represent the same real-world entity. Existing approaches tr
eat ER focus only on finding perfectly matched records and separating the
corresponding from non-corresponding ones. However\, in real-world scenari
os\, where ER is part of a more general data project\, downstream applicat
ions may not only require resolution of records that refer to the same ent
ity but may also seek to match records that share different levels of comm
onality\, relating\, for example\, to various granularity levels of the re
solution. In what follows\, we introduce the problem of multiple intents e
ntity resolution (MIER)\, an extension to the universal (single intent) ER
task. As a solution\, we propose FlexER\, utilizing contemporary solution
s to universal ER tasks to solve multiple intents entity resolution. FlexE
R addresses the problem as multi-label classification and combines intent-
based representations of record pairs using a graph convolutional network
(GCN) to improve the outcome to multiple resolution problems. A large-scal
e empirical evaluation introduces a new benchmark and\, using also three w
ell-known benchmarks\, shows that FlexER effectively solves the MIER probl
em and outperforms the state-of-the-art for a universal ER.\n\nZoom Link\n
\nhttps://technion.zoom.us/j/3800541616\n\n \;\n\n \;
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:111@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210808T143000
DTEND;TZID=Asia/Jerusalem:20210808T153000
DTSTAMP:20210804T055833Z
URL:https://dds.technion.ac.il/iemevents/interactive-reinforcement-learnin
g-with-dynamic-query-cost-2/
SUMMARY:Interactive Reinforcement Learning with Dynamic Query Cost [ \n
Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc Sveta Bikulov\n Advisors: Ofra Amir\n Where: ZOOM From
:\nTechnion\nAbstract:\n\nPeople increasingly interact with intelligent ag
ents. Some of these agents\, e.g. social robots\, need to adapt and person
alize to the person they interact with. To achieve this\, the agent should
be able to receive feedback about its actions and learn the users' prefer
ences. However\, people do not always provide such feedback. A possible ap
proach would be for the agent to ask the user for feedback for each action
it executes\, but this is not desirable since it might irritate the user.
Therefore\, the agent needs to decide when to ask for feedback\, taking i
nto consideration the costs and benefits of doing so. As an attempt to sol
ve this problem\, we formalize the problem of interactive Multi-Armed Band
its with dynamic query cost\, layout key challenges\, and analyze possible
solutions of existing methods to solve this problem. The project is done
in collaboration with Intuition Robotics.\n\nZoom Link\n\nhttps://technion
.zoom.us/j/3800541616\n\n \;\n\n \;
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:110@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210810T143000
DTEND;TZID=Asia/Jerusalem:20210810T153000
DTSTAMP:20210727T050901Z
URL:https://dds.technion.ac.il/iemevents/frank-wolfe-with-a-nearest-extrem
e-point-oracle/
SUMMARY:Frank-Wolfe with a Nearest Extreme Point Oracle [ \n Graduate S
tudent Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc Noam Wolf\n Advisors: Dan Garber\n Where: ZOOM From:\n
Technion\nAbstract:\n\nWe consider variants of the classical Frank-Wolfe a
lgorithm for constrained smooth convex minimization\, that instead of acce
ss to the standard oracle for minimizing a linear function over the feasib
le set\, have access to an oracle that can find an extreme point of the fe
asible set that is closest in Euclidean distance to a given vector. We fir
st show that for many feasible sets of interest\, such an oracle can be im
plemented with the same complexity as the standard linear optimization ora
cle. We then show that with such an oracle we can design new Frank-Wolfe v
ariants which enjoy significantly improved complexity bounds in case the s
et of optimal solutions lies in the convex hull of a subset of extreme poi
nts with small diameter (e.g.\, a low-dimensional face of a polytope). In
particular\, for many 0-1 polytopes\, under quadratic growth and strict co
mplementarity conditions\, we obtain the first linearly convergent variant
with rate that depends only on the dimension of the optimal face and not
on the ambient dimension.\n\nZoom Link\n\nhttps://technion.zoom.us/j/38005
41616\n\n \;\n\n \;
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:112@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210919T143000
DTEND;TZID=Asia/Jerusalem:20210919T153000
DTSTAMP:20210914T051318Z
URL:https://dds.technion.ac.il/iemevents/testing-hypotheses-on-a-tree-with
-fdr-control-for-the-highest-resolution-discoveries/
SUMMARY:Testing hypotheses on a tree with FDR control for the highest resol
ution discoveries [ \n Graduate Student Seminar\n Seminars\n \n
]
DESCRIPTION:By: M.Sc. Pnina Aizenberg\n Advisors: DR. Marina Bogomolov\n Wh
ere: ZOOM From:\nTechnion\nAbstract:\nSeveral applications require testing
a large number of statistical hypotheses with hierarchical tree structure
\, where the child hypotheses are more specific than their parent hypothes
es. In such cases\, rejection of a child hypothesis makes the rejection of
all its ancestor hypotheses redundant\, therefore it is natural to focus
on the highest resolution discoveries\, i.e. the outer nodes\, defined as
the discovered nodes that are not ancestors of other discoveries.\n\nWe pr
opose a hierarchical method for testing trees of hypotheses with outer nod
es false discovery rate (FDR) control\, which exploits the logical relatio
nships between the hypotheses in the tree. Our theoretical and numerical r
esults address separately testing trees of hypotheses which are induced by
hierarchical clustering of explanatory variables in a linear regression m
odel\, where the clustering is based on the correlations between the varia
bles. In this setting\, the method identifies the smallest clusters with e
vidence for containing important variables\, while controlling for the exp
ected proportion of discovered clusters with no important variables.\n\nOu
r method is compared to several competitors in a simulation study\, and is
shown to be more powerful in several settings. We illustrate the applicat
ion of the method for hierarchical variable selection on real data\, and s
how that in some cases it leads to more specific discoveries than its comp
etitors.\n\n \;\n\nZoom Link\n\nhttps://technion.zoom.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:114@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20210930T143000
DTEND;TZID=Asia/Jerusalem:20210930T153000
DTSTAMP:20210922T073835Z
URL:https://dds.technion.ac.il/iemevents/adaptive-methods-for-testing-hypo
theses-with-group-structure-while-simultaneously-controlling-several-error
-rates/
SUMMARY:Adaptive methods for testing hypotheses with group structure while
simultaneously controlling several error rates [ \n Graduate Student S
eminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Ido Griness\n Advisors: Dr. Marina Bogomolov\n Where
: ZOOM From:\nTechnion\nAbstract:\nIn many statistical applications a larg
e set of hypotheses is tested\, and the hypotheses can be naturally classi
fied into groups based on different criteria\, defined by the characterist
ics of the problem. Examples of such applications include brain imaging\,
microbiome\, and genome-wide association studies. In such settings\, it ma
y be of interest to identify groups containing signals\, for each partitio
n into groups\, with control over false discoveries. This goal was address
ed by Barber and Ramdas (2016) and Ramdas\, Barber\, Wainwright\, and Jord
an (2019) who developed the p-filter method for controlling the group-leve
l false discovery rate (FDR)\, simultaneously for all partitions.\n\nWe ad
dress the same goal\, and aim to improve the power of the p-filter method
by capturing the group structure of the hypotheses using adaptive weights.
We prove that the modified method controls the group-level FDR for each p
artition into groups under independence\, and show by simulations that it
seems to retain the control under certain forms of positive dependence. Ou
r simulation study shows that the proposed modification improves the power
of the method significantly in the settings where the signals are concent
rated within groups\, and does not result in a power loss in less favorabl
e settings. We compare the performance of the modified method to that of t
he original p-filter on real brain imaging data\, where the hypotheses are
grouped with respect to two criteria.\n\n\nLink to Zoom\n\nhttps://techni
on.zoom.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:116@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20211017T143000
DTEND;TZID=Asia/Jerusalem:20211017T153000
DTSTAMP:20211005T053831Z
URL:https://dds.technion.ac.il/iemevents/learning-discrete-structured-vari
ational-auto-encoder-using-natural-evolution-strategies/
SUMMARY:Learning Discrete Structured Variational Auto-Encoder using Natural
Evolution Strategies [ \n Graduate Student Seminar\n Seminars\n
\n ]
DESCRIPTION:By: MSc. Alon Berliner\n Advisors: Prof. Tamir Hazan\n Where: Z
OOM From:\nTechnion\nAbstract:\n\nDiscrete variational auto-encoders (VAEs
) are able to represent semantic latent spaces in generative learning. In
many real-life settings\, the discrete latent space consists of high-dimen
sional structures\, and propagating gradients through the relevant structu
res often requires enumerating over exponentialy many structures. Recently
\, various approaches were devised to propagate approximated gradients wit
hout enumerating over the space of possible structures. In this work\, we
use Natural Evolution Strategies (NES)\, a class of gradient-free black-bo
x optimization algorithms\, to learn discrete VAEs. NES algorithms are com
putationally appealing as they estimate gradients with forward pass evalua
tions only\, thus they do not require to propagate gradients through their
discrete structures. We demonstrate empirically that optimizing discrete
structured VAEs using NES is as effective as gradient-based approximations
. Lastly\, we prove NES converges for non-Lipschitz functions as appear in
discrete structured VAEs.\n\n \;\n\nZoom Link\n\nhttps://technion.zo
om.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:124@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20211024T143000
DTEND;TZID=Asia/Jerusalem:20211024T153000
DTSTAMP:20211021T073312Z
URL:https://dds.technion.ac.il/iemevents/wisdom-of-the-crowds-semi-strong-
form-efficiency-in-prediction-markets/
SUMMARY:Wisdom of the Crowds: Semi-Strong Form Efficiency in Prediction Mar
kets [ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M. Sc. Yosef Mentzer \n Advisors: Prof. Doron Kliger \n W
here: ZOOM From:\nTechnion\nAbstract:\n\nAccording to the Efficient Market
Hypothesis (EMH)\, prices fully reflect all relevant information at any p
oint in time. Much of the EMH theoretical and empirical basis has been cha
llenged\, researchers have been turning to theories including models of hu
man psychology\, and the field of behavioral finance emerged. In the last
decade\, due to some of its convenient features\, prediction markets have
been progressively employed by researchers as a test setting for the EMH a
nd for behavioral finance hypotheses.\n\nI gather historical data on predi
ction markets of soccer match results from an online betting exchange and
train a machine learning prediction model for the dynamics of prices assoc
iated with the team that scores a goal. I then devise an investment strate
gy based on the model's predictions and perform a Monte Carlo simulation t
o compare the returns of a naïve portfolio based on all goals with the re
turns of a portfolio based on the investment strategy. The results allow m
e to shed light on semi-strong form efficiency of the prediction market in
the minutes following a goal.\n\n \;\n\nZoom Link\n\nhttps://technion
.zoom.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:125@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20211107T143000
DTEND;TZID=Asia/Jerusalem:20211107T153000
DTSTAMP:20211028T055843Z
URL:https://dds.technion.ac.il/iemevents/a-theoretical-and-empirical-study
-of-the-weighted-completion-time-minimization-problem-with-capacitated-par
allel-machines/
SUMMARY:A theoretical and empirical study of the weighted completion time m
inimization problem with capacitated parallel machines [ \n Graduate S
tudent Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc Iyar Zaks\n Advisors: Izack Cohen\n Where: ZOOM From:\
nTechnion\nThe weighted completion time minimization problem for capacitat
ed parallel machines is a fundamental problem in modern cloud computing en
vironments. Because of the problem’s NP-hardness\, we study heuristic ap
proaches with provable approximation guarantees. Via a numerical study and
a developed mixed integer linear program of the problem\, we demonstrate
the performance of the suggested algorithm with respect to the optimal sol
utions and other alternative scheduling methods.\n\nZoom Link\n\nhttps://t
echnion.zoom.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:132@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20211114T143000
DTEND;TZID=Asia/Jerusalem:20211114T153000
DTSTAMP:20211111T063911Z
URL:https://dds.technion.ac.il/iemevents/canonical-correlation-analysis-fo
r-multi-trait-gwas/
SUMMARY:Canonical Correlation Analysis for Multi-Trait GWAS [ \n Gradua
te Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Lior Landau\n Advisors: Or Zuk and Marina Bogomolov \
n Where: ZOOM From:\nTechnion\nAbstract:\n\n \;\n\nIn this work\, we f
ocus on multi-trait GWAS with polygenic correlated traits. We present a ne
w method and software for detecting single nucleotide polymorphisms (SNPs)
that are associated with at least one trait by testing the association of
SNPs with data-dependent linear combinations of traits. Our method contro
ls the expectation of the proportion of discovered SNPs that are not corre
lated with any phenotype (FDR)\, and exploits the genetic and phenotypic c
orrelations in order to gain power for detecting SNPs associated with mult
iple phenotypes. The method operates by combining known ideas: first\, ide
ntify the most correlated linear combinations of phenotypes and SNPs based
on Canonical Correlation Analysis (CCA). Then\, aggregate for each SNP th
e association signal from the multiple linear trait combinations using a w
eighted Simes approach.\nWe tested several types of scenarios to check our
method's strength and robustness in a simulation study. We demonstrate th
at our proposed method controls the FDR and increases statistical power ov
er methods that are variations of ours as well as over known methods from
the literature. The proposed method was evaluated on the UK Biobank datase
t consisting of 800\,000 SNPs measured over 500\,000 individuals. This ana
lysis allowed us to identify SNPs that were associated with several phenot
ypes. Our approach discovered the highest number of significant SNPs both
when tested for associations with lipid-related phenotypes as well as with
BMI-related phenotypes.\n\n \;\n\nZoom Link\n\nhttps://technion.zoom.
us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:134@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20211121T143000
DTEND;TZID=Asia/Jerusalem:20211121T153000
DTSTAMP:20211117T120420Z
URL:https://dds.technion.ac.il/iemevents/automated-repair-of-neural-networ
ks/
SUMMARY:Automated Repair of Neural Networks [ \n Graduate Student Semin
ar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Dor Cohen\n Advisors: Prof. Ofer Strichman\n Where:
ZOOM From:\nTechnion\nAbstract:\n\n \;\n\nOver the last decade\, artif
icial Neural Networks (NNs) have been widely used in many applications inc
luding safety-critical ones\, such as autonomous systems. Hence\, it is hi
ghly important to provide guarantees that such systems work correctly. In
this work\, we exploit methods from the field of formal verification\, to
produce a correct system from a given specification. Specifically\, we int
roduce a framework for repairing an unsafe NN w.r.t. safety properties. Fu
rther\, we perform extensive experiments to demonstrate the capabilities o
f our proposed framework for generating correct NNs. To prove our method's
effectiveness\, we compare it to a naive baseline. Lastly\, we provide an
algorithm to automatically repair NNs given safety requirements.\n\n
\;\n\nZoom Link\n\nhttps://technion.zoom.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:140@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20211212T143000
DTEND;TZID=Asia/Jerusalem:20211212T143000
DTSTAMP:20211206T055421Z
URL:https://dds.technion.ac.il/iemevents/pipe-tightness-testing-in-fuel-st
orage-facilities/
SUMMARY:Pipe tightness testing in fuel storage facilities [ \n Graduate
Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M. Sc Arkadiy Haikin \n Advisors: Dr. Yefim Haim Michlin a
nd Prof. Eitan Naveh \n Where: ZOOM From:\nTechnion\nAbstract:\nThe piping
tightness tests used today in Israel are performed according to an outdat
ed API standard intended for underground piping only. Also\, as opposed to
the latest guidelines from the US Environmental Protection Agency - EPA\,
the standard defines that the pressure at the end of the test should not
be lower than a certain threshold and does not define absolute permissible
leak rate values.\nIn order to assess the detection capacity of a static
pressure test according to a standard and analyze ways to improve it\, a q
uantitative model was developed. The model takes into account various vari
ables\, including: test pressure\, gas-liquid proportion in the pipe and t
he temperature difference uncertainty between the begging and the end of t
he test.\nDuring the analysis it was found that in case of static pressure
test\, the proportion of the gases in the pipe has a significant effect o
n the detection capacity. Therefore\, a test procedure was proposed that p
rovides an estimate of the gas proportions in the pipe before the test is
performed. A short static pressure employed was found to provide a relativ
ely high detection capacity in the case of underground piping at low test
pressures only. It turned out that\, in order to meet the detection capaci
ty recommended by the EPA\, the pressure drop threshold must be lowered at
the end of the test.\nThe possibility of increasing test duration 24 hour
s was also examined\, it was found that the detection capacity can be sign
ificantly improved\, however there is a maximum pressure limit in the pipe
due to fluctuations in temperature during the test. Therefore\, a 24 hour
static pressure test is suitable for pipes that are mostly underground\,
so the average temperature in them varies very little.\nIn order to provid
e an excess pressure increase solution\, the possibility of maintaining a
constant pressure during the test was examined by addition and draining th
e test liquid from the pipe to a balanced reservoir. It was found that 24
hour test at constant pressure provides a very high detection capacity\, w
ithout the need to estimate the proportion of gases in the pipe. Moreover\
, the test is suitable for a wide range of pressures and above all is suit
able for piping with varying degree of exposure to environment. Finally\,
a conceptual design of a relatively simple test system for performance of
constant and controlled pressure test\, was presented.\n\n \;\n\nZoom
Link\nhttps://technion.zoom.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:157@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220109T143000
DTEND;TZID=Asia/Jerusalem:20220109T153000
DTSTAMP:20220104T065910Z
URL:https://dds.technion.ac.il/iemevents/conformance-checking-over-stochas
tically-known-logs/
SUMMARY:Conformance Checking over Stochastically Known Logs [ \n Gradua
te Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Eli Bogdanov\n Advisors: Dr. Izack Cohen and Prof. Av
igdor Gal\n Where: ZOOM From:\nTechnion\nAbstract:\n\nProcess mining facil
itates data-driven process modeling\, analysis and optimization by applyin
g techniques from Data Science\, Information Systems and Operations Manage
ment disciplines. The three main process mining tasks are: process discove
ry\, conformance checking and process enhancement. The data for these task
s is often stored in the form of event logs and collections of traces wher
e each trace is a sequence of events and activities that were created foll
owing a specific process realization. Some of these data are uncertain for
a variety of reasons. Data uncertainty may be attributed to technical rea
sons' such as sensor inaccuracies\, the use of probabilistic data classifi
cation models\, data quality reduction during processing and low quality o
f data capturing devices. We focus on process mining with uncertain event
data when the probability distribution functions of the event data are kno
wn. In this paper we propose a new algorithm for conformance checking in s
uch a setting\, characterize and mathematically define the building bloc
ks for stochastic conformance checking and conducting experiments in order
to evaluate our approach.\n\n \;\n\nZoom Link\n\nhttps://technion.zoo
m.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:158@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220116T143000
DTEND;TZID=Asia/Jerusalem:20220116T153000
DTSTAMP:20220105T114452Z
URL:https://dds.technion.ac.il/iemevents/inventory-optimization-in-omni-ch
annel-fulfilment-model/
SUMMARY:Inventory optimization in Omni-channel fulfilment model [ \n Gr
aduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Hagai Rettig\n Advisors: Prof. Yale Herer\n Where: ZO
OM From:\nTechnion\nAbstract:\n\nWe study a model of a supply chain chain
network\, combining traditional Brick &\; Mortar (B&\;M) stores with
collocated click &\; collect points. These points are supplied by eith
er a dark store or B&\;M store itself. For this model we investigate se
tting the optimal inventory at the beginning of every period at each B&
\;M store and the dark store. We also investigate the optimal fulfilment p
olicy when customer demand is revealed. We first find the optimal fulfilme
nt policy and then formulate the problem as a two stage stochastic problem
with recourse. We find the optimal inventory levels using Infinitesimal P
erturbation Analysis.\n\n \;\n\nZoom Link\n\nhttps://technion.zoom.us/
j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:163@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220117T113000
DTEND;TZID=Asia/Jerusalem:20220117T123000
DTSTAMP:20220110T064342Z
URL:https://dds.technion.ac.il/iemevents/a-first-order-method-for-solving-
non-differentiable-and-non-strongly-convex-bilevel-optimization/
SUMMARY:A First Order Method for Solving Non-Differentiable and Non-strongl
y convex Bilevel Optimization [ \n Graduate Student Seminar\n Semi
nars\n \n ]
DESCRIPTION:By: M.Sc. Roey Merchav\n Advisors: Prof. Shoam Sabach\n Where:
ZOOM From:\nTechnion\nAbstrasct:\n\nBi-level optimization problems seek t
o find a minimizer of an outer objective function constrained to the minim
izers set of an inner optimization problem. These problems\, both in the c
onvex and non-convex settings\, have diverse applications in the fields of
machine learning\, signal processing and many more. Since these problems
inherently don't satisfy Slater's condition\, exact solutions are mostly d
ifficult to find. Consequently\, existing algorithms only focus on approxi
mating them. Most algorithms for bi-level optimization problems are based
on regularization techniques\, which regularize the outer objective functi
on with the inner objective function in a certain way. Moreover\, in the c
onvex setting\, all the proposed algorithms provide (if any) a convergence
rate result only in terms of the inner objective function\, but only unde
r restrictive assumptions such as the outer objective function is smooth a
nd/or strongly convex (except a very recent work that propose a very compl
icated algorithm).\n\nIn this thesis\, we propose the Bi-Sub-Gradient (Bi-
SG) method\, which is a generalization of the classical sub-gradient metho
d to the setting of bi-level optimization problems. This is a first-order
method that is very easy to implement in the sense that it requires only a
computation of the associated proximal mapping or a sub-gradient of the o
uter objective function\, in addition to a proximal gradient step on the i
nner optimization problem (a step that is shared by many algorithms). We s
how\, under very mild assumptions\, that Bi-SG tackles bi-level optimizati
on problems and achieves several theoretical guarantees. First\, Bi-SG enj
oys sub-linear rates both in terms of the inner and outer objective functi
ons. Moreover\, if the outer objective function is additionally strongly c
onvex (could be still non-smooth)\, the outer rate can be improved to a li
near rate. Last\, we prove that the distance of the generated sequence to
the set of optimal solutions of the bi-level problem converges to zero.\n\
nIt should be noted that our mild assumptions do not include any different
iablity of the outer objective function nor its strong convexity. For inst
ance\, any smooth or Lipschitz continuous outer objective function satisfi
es the needed assumptions. Finally\, we demonstrate Bi-SG's performances i
n an extensive numerical comparison.\n\n \;\n\nZoom Link\n\nhttps://te
chnion.zoom.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:164@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220120T103000
DTEND;TZID=Asia/Jerusalem:20220120T113000
DTSTAMP:20220112T074447Z
URL:https://dds.technion.ac.il/iemevents/open-surgery-tool-classification-
and-hand-utilization-using-a-multi-camera-system/
SUMMARY:Open surgery tool classification and hand utilization using a multi
-camera system [ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Kristina Basiev\n Advisors: Prof. Shlomi Laufer\n Whe
re: ZOOM From:\nTechnion\nAbstract:\n\nThe goal of this work is to use mul
ti-camera video to classify open surgery tools as well as identify which
tool is held in each hand. Multi-camera systems help prevent occlusions i
n open surgery video data. Furthermore\, combining multiple views such as
a Top-view camera covering the full operative field and a Close-up camera
focusing on hand motion and anatomy\, may provide a more comprehensive v
iew of the surgical workflow. However\, multi-camera data fusion poses a
new challenge: a tool may be visible in one camera and not the other. Thu
s\, we defined the global ground truth as the tools being used regardless
their visibility. Therefore\, tools that are out of the image should be
remembered for extensive periods of time while the system responds quic
kly to changes visible in the video.\n\nZoom Link\n\nhttps://technion.zoom
.us/j/3800541616\n\n \;
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:165@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220123T143000
DTEND;TZID=Asia/Jerusalem:20220123T143000
DTSTAMP:20220116T095310Z
URL:https://dds.technion.ac.il/iemevents/using-sub-document-units-for-docu
ment-and-passage-retrieval/
SUMMARY:Using Sub Document Units for Document and Passage Retrieval [ \n
Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: Ph.D. Eilon Sheetrit\n Advisors: Prof. Oren Kurland\n Where
: ZOOM From:\nTechnion\nAbstract:\nThe ease of finding and retrieving info
rmation has become an integral part in our lives. With the growing surge o
f data\, search engines facilitate the task of finding the relevant inform
ation pertaining to a need in large collections. Yet\, searching massive v
olumes of diverse textual documents in order to satisfy a specific informa
tion need of a user is extremely challenging.\n\nThe retrieved units in co
mmercial search engines\, as is the case for enterprise engines\, are full
\ndocuments\; i.e.\, documents in a corpus are ranked by their presumed re
levance to the information need expressed by a query. Relevant documents c
an contain much non-relevant information\; specifically\, only a short pas
sage with relevant information suffices to deem the entire document releva
nt. This fact has motivated work on passage retrieval and passage-based do
cument retrieval. In the former\, a.k.a\, focused retrieval\, the retrieve
d units are passages\, and in the latter\, the retrieved units are documen
ts but passage-level information is used for document ranking.\n\nIn this
thesis\, we first present a suite of novel document retrieval methods that
are based on learning document ranking function using an effective passag
e ranking. We then explore the use of inter-passage similarities to improv
e the effectiveness of the retrieved list of passages. To lay theoretical
grounds for the use of inter-passage similarities\, we propose a novel set
of cluster hypothesis tests for passages. Finally\, we examine and analyz
e the use of true relevance feedback at the token level on the retrieval p
erformance of document ranking.\n\nZoom Link\n\nhttps://technion.zoom.us/j
/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:169@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220208T150000
DTEND;TZID=Asia/Jerusalem:20220208T160000
DTSTAMP:20220203T064125Z
URL:https://dds.technion.ac.il/iemevents/making-scientific-articles-abstra
ct-understandable-for-laypeople/
SUMMARY:Making scientific articles abstract understandable for laypeople [
\n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Sapir Friedman\n Advisors: Dr. Elad Yom-Tov\n Where:
ZOOM From:\nTechnion\nAbstract:\n\nThere is an abundance of scientific kno
wledge in academic papers\, but laypeople often find the scientific langu
age or these papers difficult to understand. In this work\, we designed
an extractive summarization algorithm which creates simplified summa
ries for the abstracts of scientific articles. We trained our algorithm
using a novel parallel corpus derived from Reddit posts and comments
that refer to a medical scientific article. We compared the proposed al
gorithm to several previously-proposed methods in terms of the faithfulne
ss of their summaries\, their quality\, readability and the level of u
nderstanding that laypeople obtained from the summaries. We show that ea
ch captures a different facet of the summaries and demonstrate that the pr
oposed algorithm achieves superior performance across the four measures.\n
\n \;\n\nZoom Link\n\nhttps://technion.zoom.us/j/3800541616
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:166@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220213T141500
DTEND;TZID=Asia/Jerusalem:20220213T151500
DTSTAMP:20220124T111158Z
URL:https://dds.technion.ac.il/iemevents/title-fairness-views-in-the-israe
li-population/
SUMMARY:Title: Fairness Views In The Israeli Population [ \n Graduate S
tudent Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Rachelle Cohen\n Advisors: Dr. Amnon Maltz\n Where:
ZOOM From:\nUniversity of Haifa\nAbstract:\nPeople's attitudes regarding t
he fairness of economic inequality may differ depending on the source of i
nequality. The fairness views with respect to income inequality generated
by luck or by merit have been recently studied in different countries. In
this work we extend the discussion and examine views regarding inherited i
nequality in addition to views regarding inequality generated by luck or b
y merit. We conduct an incentivized online experiment consisting of a repr
esentative sample of the Israeli population. The experiment is based on th
e impartial spectator design (Almas et al.\, 2020) in which subjects may r
edistribute unequal earnings of two other subjects in situations where the
initial inequality is due to luck\, effort or inheritance. First\, we rep
licate the findings by Almas et al. (2020): people redistribute more when
inequality is due to luck than when it is due to merit. Second\, redistrib
ution choices in the case of inherited inequality are similar to those mad
e when inequality is due to luck. In other words\, inherited inequality is
perceived as unfair as inequality that is generated by luck. Our findings
carry important policy implications regarding inheritance tax\, a tax tha
t has been repealed in Israel in 1980.\n\nAcademic Program: Joint Master's
Program in Economics - Specialization in Behavioral Economics\n\n \;\
n\nZoom Link\n\nhttps://us02web.zoom.us/j/87373907113
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:170@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220214T140000
DTEND;TZID=Asia/Jerusalem:20220214T140000
DTSTAMP:20220209T131140Z
URL:https://dds.technion.ac.il/iemevents/%d7%9b%d7%95%d7%aa%d7%a8%d7%aa-be
havioral-biases-on-large-scale-markets-how-do-behavioral-biases-affect-mar
kets-activity/
SUMMARY:Behavioral biases on large scale markets How do Behavioral biases
affect markets activity? [ \n Graduate Student Seminar\n Seminars\
n \n ]
DESCRIPTION:By: M.Sc. Lior Bakalo\n Advisors: Prof Todd Kaplan \n Where: ZO
OM From:\nTechnion and Haifa University\nAbstract:\n\nWe assume markets to
be efficient\, truthfully reflecting capital information. On the other ha
nd\, those markets are composed of biased individuals. Our paper focuses o
n sports betting markets\, with the purpose of creating the linkage betwee
n individuals' decision making process and market activity. Doing so\, we
provide a novel method used to extract the elasticity of demand out of con
tingent claim markets. Furthermore\, we introduce new insights on the bigg
est known market bias - The Long-Shot Bias. Our findings suggest that mark
ets and prices are highly affected by individuals' biases.\n\nZoom Link\n\
nhttps://us02web.zoom.us/j/85074250285
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:171@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220310T140000
DTEND;TZID=Asia/Jerusalem:20220310T150000
DTSTAMP:20220214T105122Z
URL:https://dds.technion.ac.il/iemevents/on-the-global-convergence-of-mult
idimensional-scaling/
SUMMARY:On the Global Convergence of Multidimensional Scaling [ \n Grad
uate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Noga Ram\n Advisors: Associate Prof. Sabach Shoham\n
Where: ZOOM From:\nTechnion\nAbstract:\n\nMultidimensional Scaling (MDS) i
s a popular tool for dimensionality reduction and data visualization. From
image processing to stock-market analysis and all the way to medical diag
nosis\, MDS is a powerful technique for extracting meaningful insights out
of data. In this talk\, we focus on the most widely used approach to form
ulate the MDS problem\, which results in a challenging non-smooth and non-
convex optimization problem. We propose the first globally convergent iter
ative MDS algorithm\, along with a simple inertial-based acceleration sche
me (which is also proven to globally converge).\n\n \;\n\nZoom Link\n\
nhttps://technion.zoom.us/j/93063946490
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:174@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220320T163000
DTEND;TZID=Asia/Jerusalem:20220320T173000
DTSTAMP:20220313T053023Z
URL:https://dds.technion.ac.il/iemevents/less-is-more-phenomenon-in-superv
ised-machine-learning/
SUMMARY:Less is More" Phenomenon in Supervised Machine Learning [ \n Gr
aduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Alexander Chapanin\n Advisors: Prof. Carmel Domshlak\
n Where: Bloomfield 424 From:\nTechnion\nSupervised function learning is
performed under an assumption that the training examples are sampled i.i.d
\, faithfully representing the data distribution\, and thus the larger the
training sample is\, the better. It was observed\, however\, that careful
ly selected non-i.i.d. training sets may reduce the amount of training dat
a needed to achieve quality learning.\n\n \;\n\nThis fascinating ``les
s is more" phenomenon has been so far left largely unexplained and this is
precisely what we investigate in our work here. First\, we examine the tw
o hypotheses for the source of the ``less is more" phenomenon that has bee
n proposed in previous works. We then propose a different theory for its e
mergence\, and empirically validate our theory on a comprehensive set of b
enchmarks. Based on our findings\, we then propose several simple algorith
ms for incremental learning that aim at focusing on the right subset of th
e training data\, and empirically validate and demonstrate their effective
ness.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:177@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220322T100000
DTEND;TZID=Asia/Jerusalem:20220322T110000
DTSTAMP:20220315T092853Z
URL:https://dds.technion.ac.il/iemevents/infectious-disease-households-mod
eling-with-missing-data/
SUMMARY:Infectious Disease: Households Modeling with Missing Data [ \n
Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Oron Madmon\n Advisors: Prof. Yair Goldberg\n Where:
Conference room at the statistics laboratory\, Cooper building\, entrance
floor From:\nTechnion\nAbstract:\n\nOver two years after the first identi
fied SARS-CoV-2 case\, the role of adolescents and children in spreading t
he virus remains unclear. In our work\, we generalize a well-known househo
ld model for modeling infectious diseases\, to include missing tests. Due
to missingness\, the likelihood of the generalized model cannot be evaluat
ed explicitly\, and so does the MLE. Thus\, we propose an estimation metho
dology\, using a novel EM algorithm\, for estimating the MLE in the presen
ce of missing data. We implement the proposed mechanism using R software.
We illustrate\, using a simulation study\, the performance of the proposed
estimation methodology\, in comparison with the estimation procedure in t
he complete case. Finally\, using the proposed estimation methodology we a
nalyzed a dataset containing SARS-CoV-2 testing results\, collected from t
he city of Bnei Brak\, Israel\, during the beginning of the pandemic.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:178@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220327T163000
DTEND;TZID=Asia/Jerusalem:20220327T173000
DTSTAMP:20220315T093206Z
URL:https://dds.technion.ac.il/iemevents/hi-c-based-haplotype-phasing-with
-machine-learning/
SUMMARY:Hi-C-based haplotype phasing with machine learning [ \n Graduat
e Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Aviv Zeilig\n Advisors: Ass. Prof. Noam Kaplan \n Whe
re: Bloomfield 424 From:\nTechnion\nAbstract:\n\nHaplotype phasing is the
process of differentiating genetic variations between two homologous chrom
osomes and is traditionally inferred statistically from a population or de
termined by long-range sequencing of an individual genome. However\, exten
ding haplotype inference to the whole-chromosome scale remains challenging
. Here we propose a general ML strategy to determine complete chromosomal
haplotypes using Hi-C data. Due to the patterns that emerge from Hi-C maps
the data can be formulated as a clustering problem. The underperformance
of standard generic clustering algorithms on noisy high-dimensional biolog
ical data has led to the development of KMD clustering a generic clusterin
g approach\, based on a simple generalization of single and average linkag
e hierarchical clustering. When compared to standard generic and state-of-
the-art specialized algorithms\, KMD clustering’s performance was consis
tently better or comparable to that of the best algorithm on each of the t
ested datasets.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:192@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220410T163000
DTEND;TZID=Asia/Jerusalem:20220410T173000
DTSTAMP:20220410T075956Z
URL:https://dds.technion.ac.il/iemevents/the-association-between-the-compl
eteness-of-surgical-safety-checklists-with-safety-events-and-patient-outco
mes-a-case-control-study/
SUMMARY:The association between the completeness of Surgical Safety Checkli
sts with safety events and patient outcomes: A case-control study [ \n
Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Lior Drucker\n Advisors: Prof. Eitan Naveh\n Where: B
loomfield 424 From:\nTechnion\nAbstract:\nAccording to the Joint Commissi
on on Accreditation of Healthcare Organizations (JCAHO)\, operation room m
istakes such as wrong patient\, procedure\, or surgery site\, and lack of
necessary equipment\, represent a significant cause of surgical complicati
ons and mortality\, which are unexplained by the underlying morbidity.\nIn
the last decade and as part of improving patient safety\, the operation r
oom Surgical Safety Checklists (SSCs) have been increasingly used. The mos
t implemented is the “Time-Out Checklist” just before the surgery begi
ns. The question of whether SSCs control enhances safety culture\, decreas
es the odds for safety events\, and improves patient outcomes\, is still d
ebated.\nOur study evaluates through a case-control study the association
between the completeness of SSCs\, safety events\, and patient outcomes. W
e show that impaired patient and surgery data authentication is more likel
y to cause safety events or medical mistakes. However\, patient preparedne
ss and equipment preparedness were not associated with a safety event or a
medical mistake. This talk focuses on the “best” predictors (remained
by our model) of safety events. In addition\, it illustrates the relation
ships between safety events and patient outcomes measures (surgical compli
cations\, length of hospital stay\, and readmissions) through multivariate
logistic regression models.\nWe conducted this study in cooperation with
an Israeli single tertiary hospital.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:193@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220427T100000
DTEND;TZID=Asia/Jerusalem:20220427T110000
DTSTAMP:20220403T064256Z
URL:https://dds.technion.ac.il/iemevents/timely-and-accurate-prediction-of
-global-anomalies/
SUMMARY:Timely and accurate prediction of global anomalies [ \n Graduat
e Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Noam Arbel\n Advisors: Assoc. Prof. Erez Karpas\n Whe
re: Cognitive Robotics Lab From:\nTechnion\nAbstract:\n\nTemperatures or
pressures that are too high or too low can cause issues in the food manufa
cturing process\, such as problems with the food formation and the crystal
lization process. This type of event is referred to as a global anomaly\,
but more generally\, it can be thought of as a point of data that exceeds
a predefined set of bounds.\n\nEarly detection of such issues can help pre
vent food waste and stoppages of the manufacturing process which could lea
d to additional penalties. While it is desirable to detect such issues as
early as possible\, there is an expected trade-off between the prediction
horizon and the accuracy of the model.\n\nWe aim to predict global anomali
es while balancing the aforementioned trade-off. To do so\, we model the d
ecision process as a Markov Decision Process (MDP)\, while at any time poi
nt\, our algorithm indicates if an alert should be raised. An alert sugges
ts that a global anomaly will occur within the prediction horizon.\n\nWe u
se the Trial-based Heuristic Tree Search (THTS) framework to solve the MDP
. In addition\, the sensory data is used by a forecasting model which is f
urther used as the environment of the tree search.\n\nWe evaluate the perf
ormance of our algorithm using a synthetic dataset and compare it to basel
ines models which use the same forecasting model as our algorithm use. Our
evaluation metric is the reward function of the MDP\, which considers bot
h accuracy and timeliness.\n\n \;
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:200@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220501T163000
DTEND;TZID=Asia/Jerusalem:20220501T173000
DTSTAMP:20220425T063209Z
URL:https://dds.technion.ac.il/iemevents/uncertainty-in-service-systems-pe
rformance-measure-estimation-and-optimization-methods-for-contact-centers-
with-information-uncertainty/
SUMMARY:Uncertainty in Service Systems: Performance Measure Estimation and
Optimization Methods for Contact Centers with Information Uncertainty [ \n
Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: Ph.D. Antonio Castellanos\n Advisors: Assoc. Prof. Galit Yo
m-Tov\n Where: Bloomfield 424 From:\nTechnion\nAbstract:\nContact centers
are growing in their use more and more due to the economic value they pro
vide companies\, and more so during the COVID-19 pandemic\, where social d
istancing became a constraint.\nContact centers are rich environments with
new types of data that enable us researchers to deepen the understanding
of the way that service is created and managed. Specifically\, data from c
ontact centers includes information on how the interaction between the par
ties evolved. In my PhD I investigated such data from two service channels
: chat and in-app messaging systems\, which LivePerson Inc. made available
to the SEElab at the Technion.\nAnalyzing the data of these two service c
hannels\, I recognized that the usual ways of analyzing and measuring qual
ity in call centers give biased estimations of performance levels when app
lied to contact centers. The reason for the biased measures is the way cus
tomers and employees behave in these systems\, which creates various types
of information uncertainty. For example\, it is uncertain how many custom
ers are waiting in the queue\, since some of the customers that abandon th
e queue do so without closing the communication window or the dedicated ap
plication. In addition\, it is uncertain how many customers are currently
in service\, because employees do not close the communication window immed
iately upon service completion since there is some probability that the cu
stomer might write an additional message. The uncertainty that exists in c
ontact centers calls for new mathematical modeling that will enable compan
ies to make better operational decisions under such uncertainty. Consequen
tly\, the main goal of my PhD research was to develop stochastic models to
elevate operational decision making in contact centers. To achieve this\,
I combined methodologies from the data science\, statistical and optimiza
tion literatures.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:202@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220508T163000
DTEND;TZID=Asia/Jerusalem:20220508T173000
DTSTAMP:20220426T025029Z
URL:https://dds.technion.ac.il/iemevents/impact-of-procedural-and-distribu
tive-justice-on-patient-flow-in-hospitals/
SUMMARY:Impact of Procedural and Distributive Justice on Patient flow in Ho
spitals [ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: MSc. Matias Kohn\n Advisors: Assoc. Prof. Galit Yom-Tov\n
Where: Bloomfield 424 From:\nTechnion\nAbstract:\n\nWe investigate the ope
rational impact of procedural and distributive justice in healthcare syste
ms. Specifically\, we analyze changes in routing procedures implemented in
an Israeli hospital on patient length of stay (LOS). The new routing proc
edure routes patients between ED and inpatient wards using a round-robin a
lgorithm instead of according to beds’ availability.\n\nAfter interviewi
ng medical personnel\, we identify perceived improvement in justice and fa
irness. Medical personnel report on higher control over patient discharges
and transfers between units as well as increased motivation.\n\nUsing dif
f-in-diff analysis we show a reduction of 14.6% (~1.1 hours) in hospitaliz
ation time for internal ward patients. We also show a reduction of 17.4% (
~2.1 hours) in patients’ LOS in the emergency department after implement
ing round-robin routing for hospitalized patients. We investigate the mech
anisms that drive this reduction. Specifically\, we analyze how the new ro
uting policy improved fairness by balancing the proportion of overall and
elderly patients between wards.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:209@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220522T163000
DTEND;TZID=Asia/Jerusalem:20220522T173000
DTSTAMP:20220510T070804Z
URL:https://dds.technion.ac.il/iemevents/improving-completeness-of-regress
ion-verification/
SUMMARY:Improving Completeness of Regression Verification [ \n Graduate
Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: MSc. Chaked Roger Joseph Sayedoff\n Advisors: Prof. Ofer S
trichman\n Where: Bloomfield 527 From:\nTechnion\nAbstract:\nRVT is a regr
ession verification tool for proving partial equivalance for pairs of prog
rams\, i.e.\, that the two functions emit the same output if they are fed
with the same input and they both terminate. To do so\, RVT traverses bott
om-up on the call graphs of the pair\, turn loops into recursions\, abstra
ct the recursive calls with uninterpreted functions and abstract pairs tha
t are proved as equivalent with uninterpreted functions. This enables it t
o create verification conditions in the form of small programs that are lo
op- and recursion-free. This method works well as long as the two compared
recursions are in sync. In this work we study the problem of proving equi
valence when the two recursive functions are not in sync. We extend previo
us work that studied this problem for functions with a single recursive ca
ll to the general case. We also introduce a method for detecting automatic
ally the unrolling that is necessary for making two recursive functions sy
nchronize\, when possible.\n\n \;
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:205@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220529T163000
DTEND;TZID=Asia/Jerusalem:20220529T173000
DTSTAMP:20220503T111131Z
URL:https://dds.technion.ac.il/iemevents/national-parks-operations-in-the-
covid-era-balancing-accessibility-and-overcrowding-using-online-reservatio
n-systems/
SUMMARY:National Parks Operations in the COVID Era: Balancing Accessibility
and Overcrowding using Online Reservation Systems [ \n Graduate Stude
nt Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Yamit Leon\n Advisors: Assoc. Prof. Galit Yom-Tov\n W
here: Bloomfield 527 From:\nTechnion\nAbstract:\nManagement of national pa
rks requires balancing a tradeoff between protecting nature and its inhabi
tants\nby preventing or limiting human access to parks and increasing awar
eness to its wonders by allowing people\nto access its trails and public a
reas. For many years the common method for balancing the two goals was\nto
limit humans access in national parks to specific trails and visiting hou
rs. During 2019 the Israel Nature\nand Parks Authority (INPA) started to c
ontrol access to its national parks through a requirement to set an\nappoi
ntment before visiting. This initiative started due to the COVID-19 pandem
ic regulations that raised\nthe need to control crowdedness. Yet\, the INP
A still use this system to balance and control load in popular\nparks.\nSu
ch appointment system needs to set a) the amount of daily visitors that sh
ould be allowed to visit the\npark\, and b) the number of appointment slot
s that the system will open before the day of arrival. The two\nare not id
entical\, due to the phenomena of no-show and appointment cancellation.\nI
n this research\, we develop two operational models. The first model deter
mines the optimal number of\ndaily visitors that balance crowdedness and a
ccessibility costs. The second model defines a dynamic policy\nfor the num
ber of open slots in the appointment book\, for the days before arrival. W
e study data provided\nby the INPA\, to understand visiting demand and peo
ple’s no-show and cancellation behavior. We implement\nthe developed mod
els to data of two specific parks.\nThe models we develop can be used to c
ontrol load in other leisure industries\, such as campsites\, theme\nparks
\, and museums.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:204@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220612T163000
DTEND;TZID=Asia/Jerusalem:20220612T173000
DTSTAMP:20220502T054736Z
URL:https://dds.technion.ac.il/iemevents/revisiting-the-influence-of-workl
oad-on-error-occurrence-using-data-collected-through-sensors-technology-a-
case-of-cancer-ambulatory-hospital-nurses-adaptive-behavior/
SUMMARY:Revisiting the influence of workload on error occurrence using data
collected through sensors technology: A case of Cancer Ambulatory hospita
l nurses’ adaptive behavior [ \n Graduate Student Seminar\n Se
minars\n \n ]
DESCRIPTION:By: Ph.D. Noa Nissinboim\n Advisors: Professor Eitan Naveh\n Wh
ere: Bloomfield 527 From:\nTechnion\nAbstract:\nErrors are an everyday co
ncern in organizations\, particularly in hospitals\, in which errors are o
ccasionally argued to be a result of workload. In this research we aim to
improve our understanding of the relationships between workload and errors
. We explore a counterintuitive case of a decrease in error rates when wor
kload is heavy. We drew on sensor technologies used to locate individuals\
, that provides an opportunity to generate and test alternative potential
explanation. Our results are based on data collected by 1000 sensors every
three seconds for more than two years of real-time location in 6 oncology
infusion units in one ambulatory hospital. We integrated the sensors’ d
ata set with data set of patients’ scheduled appointments\, and with dat
a set of error reports. Integration of the three data sets allowed us to s
how that nurses’ positive adaptive behavior during heavy workload circum
stances leads to a valid process improvement that decreased errors in heav
y workload situations.\n\n \;
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:210@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220619T163000
DTEND;TZID=Asia/Jerusalem:20220619T173000
DTSTAMP:20220510T071052Z
URL:https://dds.technion.ac.il/iemevents/explainable-reinforcement-learnin
g-through-integration-of-policy-summaries-and-reward-decomposition/
SUMMARY:Explainable Reinforcement Learning through Integration of Policy Su
mmaries and Reward Decomposition [ \n Graduate Student Seminar\n S
eminars\n \n ]
DESCRIPTION:By: MSc. Yael Friedler\n Advisors: Dr. Ofra Amir\n Where: Bloo
mfield 527 From:\nTechnion\nAbstract:\nExplaining the behavior of agents o
perating in sequential decision-making settings is challenging\, as their
behavior is affected by a dynamic environment and delayed reward. In this
paper\, we study a new way of combining local and global explanations of s
equential decision-making agents in order to help understand their behavio
r. Specifically\, we combine reward decomposition\, a local explanation me
thod that exposes agent preferences\, with HIGHLIGHTS\, a global explanati
on method that shows a summary of the agent's behavior in ``important'' st
ates. We conducted a user study to evaluate the integration of these expla
nation methods and their respective benefits. Our results show that local
information in the form of reward decomposition contributed to participant
s' understanding of agents' preferences\, while HIGHLIGHTS summaries did n
ot lead to an improvement compared to a baseline showing frequent agent tr
ajectories.\n\n \;
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:232@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220623T163000
DTEND;TZID=Asia/Jerusalem:20220623T173000
DTSTAMP:20220807T124802Z
URL:https://dds.technion.ac.il/iemevents/facilitate-asynchronous-data-scie
nce-invention-activities-at-scale/
SUMMARY:Facilitate Asynchronous Data Science Invention Activities at Scale
[ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: PhD Rafael Shalala\n Advisors: Assistant Prof. Ofra Amir &
Associate Prof. Ido Roll\n Where: Bloomfield 527 From:\nTechnion\nAbstract
:\n\nInvention activities are carefully designed problem-solving tasks in
which learners are asked to invent solutions to unfamiliar problems prior
to being taught the canonical solutions. Invention activities are typicall
y used in the classroom setting. As online education becomes increasingly
common\, there is a need to adapt Invention activities to the asynchronous
nature and facilitate their delivery and analysis in a larger scale. We s
tart by focusing on the invention process itself and its outcomes based on
a case study in which we analyze video recordings we collected of several
students who worked on these activities in pairs as part of an introducto
ry undergraduate data science course. We discuss lessons learned and impli
cations for the design of asynchronous Data science Invention activities.
Then we focus on facilitating these activities at scale based on a second
case study we ran in the following year in which we test the delivery and
submission of the activities\, and present analysis of the activities usin
g a dedicated framework. The framework serves as an intelligent submission
system to support scalability while also providing instant personalized f
eedback to the students to address challenges raised from the asynchronous
nature of the activities. We use the framework to analyze the solutions a
nd validate the efficacy of the activities at scale.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:233@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220626T163000
DTEND;TZID=Asia/Jerusalem:20220626T173000
DTSTAMP:20220807T124954Z
URL:https://dds.technion.ac.il/iemevents/stochastic-alternating-directions
-method-of-multipliers-for-composite-linear-optimization/
SUMMARY:Stochastic Alternating Directions Method of Multipliers for Composi
te Linear Optimization [ \n Graduate Student Seminar\n Seminars\n
\n ]
DESCRIPTION:By: M.Sc. Dan Greenstein\n Advisors: Assis. Prof. Nadav Hallak
\n Where: Bloomfield 527 From:\nTechnion\nAbstract:\n\nWe consider the min
imization of a sum of a smooth function with a nonsmooth composite functio
n\, where the composition is applied on a random linear mapping. This rand
om composite model encompasses many problems\, and can especially capture
realistic scenarios in which the data is sampled during the optimization p
rocess. We propose and analyze a method that combines the classical Augmen
ted Lagrangian framework with a sampling mechanism and adaptive update of
the penalty parameter. We show that every accumulation point of the sequen
ce produced by our algorithm is a critical point..
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:224@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220626T173000
DTEND;TZID=Asia/Jerusalem:20220626T183000
DTSTAMP:20220601T091058Z
URL:https://dds.technion.ac.il/iemevents/combinatorial-methods-for-designi
ng-observational-study-with-two-control-groups/
SUMMARY:Combinatorial Methods for Designing Observational Study with Two Co
ntrol Groups [ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Yu Chen\n Advisors: Prof. Asaf Levin\n Where: Bloomfi
eld 527 From:\nTechnion\nAbstract:\nObservational study is proposed to dea
l with the independent variables that are not under the control of the res
earchers. The problem we consider here is: there is one experiment group E
and two disjoint controls groups C_1 and C_2. These control groups are al
so disjoint from E. And there is a nonnegative weight between every pair o
f elements of distinct groups. The goal is to find a cover of E and subset
s of C_1 and C_2 through matching such that each match consists of one ele
ment of the experiment group E\, k_1 elements of the first control group C
_1 and k_2 elements of the second control group C_2\, maximizing the total
similarity where k_1 and k_2 are two parameters of the problem.\nIn this
research\, we first prove that the problem is as least as hard as to appro
ximate the densest k-subgraph problem for k_1 = k_2 = k for large constant
k. Then\, we develop several approximation algorithms with good approxima
tion ratios for k_1 = k_2 = 1 based on combinatorial methods.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:234@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220627T123000
DTEND;TZID=Asia/Jerusalem:20220627T133000
DTSTAMP:20220807T125210Z
URL:https://dds.technion.ac.il/iemevents/combinatorial-methods-for-designi
ng-observational-study-with-two-control-groups-2/
SUMMARY:Combinatorial Methods for Designing Observational Study with Two Co
ntrol Groups [ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Yu Chen\n Advisors: Prof. Asaf Levin\n Where: Bloomfi
eld 153 From:\nTechnion\nAbstract:\nObservational study is proposed to dea
l with the independent variables that are not under the control of the res
earchers. The problem we consider here is: there is one experiment group E
and two disjoint controls groups C_1 and C_2. These control groups are al
so disjoint from E. And there is a nonnegative weight between every pair o
f elements of distinct groups. The goal is to find a cover of E and subset
s of C_1 and C_2 through matching such that each match consists of one ele
ment of the experiment group E\, k_1 elements of the first control group C
_1 and k_2 elements of the second control group C_2\, maximizing the total
similarity where k_1 and k_2 are two parameters of the problem.\nIn this
research\, we first prove that the problem is as least as hard as to appro
ximate the densest k-subgraph problem for k_1 = k_2 = k for large constant
k. Then\, we develop several approximation algorithms with good approxima
tion ratios for k_1 = k_2 = 1 based on combinatorial methods.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:235@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220627T160000
DTEND;TZID=Asia/Jerusalem:20220627T163000
DTSTAMP:20220807T125725Z
URL:https://dds.technion.ac.il/iemevents/better-equilibrium-by-heterogenei
ty/
SUMMARY:Pareto efficient equilibrium selection in games [ \n Graduate S
tudent Seminar\n Seminars\n \n ]
DESCRIPTION:By: Ph.D. Gal Danino\n Advisors: Assoc. Prof. Itai Arieli \n W
here: Bloomfield 153 From:\nTechnion\nAbstract:\n\nWe consider the class o
f 2x2 coordination games where one pure equilibrium is Pareto efficient an
d the other pure equilibrium is risk dominant. We study learning dynamics
with a finite population of rational agents that take an action when they
are born in order to maximize their discounted payoffs throughout their li
fe span. The agents differ from one another by their death rate which is d
etermined according to a stochastic process. We identify a number of novel
conditions over the stochastic process for which the Pareto efficient equ
ilibrium is both accessible and absorbing and thus uniquely selected.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:236@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220627T163000
DTEND;TZID=Asia/Jerusalem:20220627T170000
DTSTAMP:20220807T130000Z
URL:https://dds.technion.ac.il/iemevents/multi-party-computation-with-priv
acy-aware-agents/
SUMMARY:Multi-Party Computation with Privacy Aware Agents [ \n Graduate
Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Roy Shahmoon\n Advisors: Prof. Rann Smorodinsky and
Prof. Moshe Tennenholtz\n Where: Bloomfield 153 From:\nTechnion\nAbstract:
\n\nA data curator would like to collect data from privacy aware agents. T
he collected data will be used for the benefit of all agents. Can the cura
tor incentivize the agents to share their data truthfully? Can he guarante
e that truthful sharing will be the unique equilibrium? Can he provide som
e stability guarantees on such equilibrium? We study necessary and suffici
ent conditions for these questions to be answered positively and complemen
t these results with corresponding data collection protocols for the curat
or. Our results account for a broad interpretation of the notion of privac
y awareness.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:246@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220727T103000
DTEND;TZID=Asia/Jerusalem:20220727T113000
DTSTAMP:20220809T110004Z
URL:https://dds.technion.ac.il/iemevents/understanding-natural-language-in
-context/
SUMMARY:Understanding Natural Language in Context [ \n Graduate Student
Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc Avichai Levy\n Advisors: Assoc. Prof. Erez Karpas\n Wh
ere: Cognitive Robotics Lab From:\nTechnion\nAbstract:\nRecent years have
seen an increasing number of applications that have a Natural Language in
terface\, either in the form of\nchatbots or via personal assistants such
as Alexa (Amazon)\, Google Assistant\, Siri (Apple)\, and Cortana (Microso
ft).\nTo use these applications\, a basic dialog between the robot and the
human is required.\nWhile this kind of dialog exists today mainly within
”static” robots that do not make any movement in the household space\,
\nthe challenge of reasoning about the information conveyed by the environ
ment increases significantly when dealing\nwith robots that can move and m
anipulate objects in our home environment.\nIn this paper\, we focus on co
gnitive robots\, which have some knowledge-based models of the world and o
perate by reasoning and planning with this model.\nThus\, when the robot a
nd the human communicate\, there is already some formalism they can use
– the robot’s knowledge representation formalism.\nOur goal in this re
search is to translate Natural Language utterances into this robot’s for
malism\, allowing much more complicated household tasks to be completed.\n
We do so by combining off-the-shelf SOTA language models\, planning tools\
, and the robot’s knowledge-base for better communication.\nIn addition\
, we analyze different directive types and illustrate the contribution of
the world’s context to the translation process.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:250@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220816T140000
DTEND;TZID=Asia/Jerusalem:20220816T144000
DTSTAMP:20220809T114605Z
URL:https://dds.technion.ac.il/iemevents/%d7%9brank-wolfe-based-algorithms
/
SUMMARY:Frank-Wolfe-based Algorithms for Approximating Tyler’s M-estimato
r [ \n Computational Data Science (CDS) Seminar\n Graduate Student
Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Lior Danon\n Advisors: Assis. Prof. Dan Graber\n Wher
e: Bloomfield 424 From:\nTechnion\nTyler's M-estimator is a well known pro
cedure for robust and heavy-tailed covariance estimation. Tyler himself su
ggested an iterative fixed-point algorithm for computing his estimator h
owever\, it requires super-linear (in the size of the data) runtime per it
eration\, which maybe prohibitive in large scale. In this work we propose\
, to the best of our knowledge\, the first Frank-Wolfe-based algorithms fo
r computing Tyler's estimator. One variant uses standard Frank-Wolfe steps
\, the second also considers away-steps (AFW)\, and the third is a \\texti
t{geodesic} version of AFW (GAFW). AFW provably requires\, up to a log fac
tor\, only linear time per iteration\, while GAFW runs in linear time (up
to a log factor) in a large n (number of data-points) regime. All three
variants are shown to provably converge to the optimal solution with subli
near rate\, under standard assumptions\, despite the fact that the underly
ing optimization problem is not convex nor smooth. Under an additional fai
rly mild assumption\, that holds with probability 1 when the (normalized)
data-points are i.i.d. samples from a continuous distribution supported on
the entire unit sphere\, AFW and GAFW are proved to converge with linear
rates. Importantly\, all three variants are parameter-free and use adapt
ive step-sizes.
CATEGORIES:Computational Data Science (CDS) Seminar,Graduate Student
Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:252@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220816T145000
DTEND;TZID=Asia/Jerusalem:20220816T153000
DTSTAMP:20220809T114539Z
URL:https://dds.technion.ac.il/iemevents/local-linear-convergence-of-gradi
ent-methods-for-subspace-optimization-via-strict-complementarity/
SUMMARY:Local Linear Convergence of Gradient Methods for Subspace Optimiza
tion via Strict Complementarity [ \n Computational Data Science (CDS)
Seminar\n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Ron Fisher\n Advisors: Assis. Prof. Dan Graber\n Wher
e: Bloomfield 424 From:\nTechnion\nAbstract: We consider optimization prob
lems in which the goal is find a k-dimensionalsubspace of the reals n-tupl
e space such that k<\;<\;n \, which minimizes a convex and smooth loss
. Such problems generalize the fundamental task of principal component ana
lysis (PCA) to include robust and sparse ounterparts\, and logistic PCA fo
r binary data\, among others. While this problem is not convex it admits n
atural algorithms with very efficient iterations and memory requirements\,
which is highly desired in high-dimensional regimes however\, arguing abo
ut their fast convergence to a global optimal solution is difficult. On th
e other hand\, there exists a simple convex relaxation for which converge
nce to the global optimum is straightforward\, however corresponding algor
ithms are not efficient when the dimension is very large. In this work we
present a natural deterministic sufficient condition so that the optimal s
olution to the convex relaxation is unique and is also the optimal solutio
n to the original nonconvex problem. Mainly\, we prove that under this con
dition\, a natural highly-efficient nonconvex gradient method\, which we r
efer to as “gradient orthogonal iteration” \, when initialized with a
``warm-start'\;'\;\, converges linearly for the nonconvex problem. W
e also establish similar results for the nonconvex projected gradient meth
od\, and the Frank-Wolfe method when applied to the convex relaxation. We
conclude with empirical evidence on synthetic data which demonstrate the a
ppeal of our approach.
CATEGORIES:Computational Data Science (CDS) Seminar,Graduate Student
Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:249@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220816T154000
DTEND;TZID=Asia/Jerusalem:20220816T162000
DTSTAMP:20220809T114240Z
URL:https://dds.technion.ac.il/iemevents/weak_oracle_based_augmented_lagra
ngian_method/
SUMMARY:Weak Oracle Based Augmented Lagrangian Method For Composite Optimiz
ation [ \n Computational Data Science (CDS) Seminar\n Graduate Stu
dent Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Tsur Livney\n Advisors: Assis. Prof. Dan Graber and A
ssoc. Prof. Shoham Sabach\n Where: Bloomfield 424 From:\nTechnion\nAbstrac
t : This paper considers a convex composite optimization problem with aff
ine constraints\, which includes problems of minimization over an in- ters
ection of convex sets. We propose an augmented Lagrangian based method\, i
n which we perform primal updates using a Weak Proximal Or- acle (WPO). Th
e WPO is an oracle more powerful than the standard linear minimization ora
cle (lmo) used in conditional gradient based meth- ods\, yet computational
ly feasible for large scale problems in interesting and important domains
such as polytopes and trace norm ball\, where the optimal solution is of l
ow rank\, in contrast to the standard quadratic min- imization oracle used
in proximal methods. For polytopes\, we show an implementation of such or
acle that requires one call for an lmo. For trace norm regularization\, as
suming the optimal solution is of low rank k\, we show that such oracle ca
n be implemented in roughly k times the com- plexity of an lmo. We also sh
ow an extension of the latter for tensors of low rank. Under an assumption
of primal quadratic gap\, we achieve convergence rate of O(1/N) on both t
he objective residual and the feasibily gap.
CATEGORIES:Computational Data Science (CDS) Seminar,Graduate Student
Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:254@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220830T113000
DTEND;TZID=Asia/Jerusalem:20220830T123000
DTSTAMP:20220821T113005Z
URL:https://dds.technion.ac.il/iemevents/stationary-hastings-levitov-in-a-
cylinder/
SUMMARY:Stationary Hastings-Levitov in a Cylinder [ \n Faculty\n Gr
aduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Anna Zhuchenko\n Advisors: Prof. Eviatar Procaccia\n
Where: Cooper 214 From:\nTechnion\nAbstract: In this research\, we study t
he Stationary Hastings-Levitov (SHL) process in a cylinder\, which represe
nts the off-lattice variant of the Diffusion Limited Aggregation (DLA) mod
el grown from a real line segment. Firstly\, we will show that the sequenc
e of growing processes defined on a cylinder converges to the SHL process
as the cylinder size\, i.e.\, radius\, goes to infinity while providing a
needed scaling for the particle size. Then\, we calculate the rate of the
process growth and infer that it is linear according to the cylinder size.
\n\n \;
CATEGORIES:Faculty,Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:256@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220912T100000
DTEND;TZID=Asia/Jerusalem:20220912T104500
DTSTAMP:20220904T055749Z
URL:https://dds.technion.ac.il/iemevents/a-zero-estimator-approach-for-est
imating-the-signal-level-in-a-high-dimensional-regression-setting/
SUMMARY:A zero-estimator approach for estimating the signal level in a high
-dimensional regression setting [ \n Graduate Student Seminar\n Se
minars\n \n ]
DESCRIPTION:By: PhD Ilan Livne\n Advisors: Prof. Yair Goldberg\n Where: Con
ference room at the statistics laboratory\, Cooper building\, entrance flo
or From:\nTechnion\nAbstract:\n\nWe study a high-dimensional linear regres
sion model in a semi-supervised setting\, where for many observations only
the vector of covariates X is given with no responses Y. We do not make
any sparsity assumptions on the vector of coefficients\, nor do we assume
normality of the covariates. We aim at estimating the signal level\, i.e.
\, the amount of variation in the response that can be explained by the se
t of covariates. We propose an estimator\, which is unbiased\, consisten
t\, and asymptotically normal. This estimator can be improved by using a z
ero-estimator approach\, where a zero-estimator is a statistic arising fro
m the unlabeled data\, whose expected value is zero. More generally\, we
present an algorithm based on the zero-estimator approach that in princip
le can improve any given estimator. We further relax the linearity assumpt
ion\, study some asymptotic properties of the proposed estimators\, and de
monstrate their finite sample performance in simulated and real datasets.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:257@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220912T104500
DTEND;TZID=Asia/Jerusalem:20220912T113000
DTSTAMP:20220904T055906Z
URL:https://dds.technion.ac.il/iemevents/pac-bayes-generalization-inequali
ties-with-data-dependent-priors-and-effect-size-quantification-for-interru
pted-time-series-analysis-with-application-to-covid-19-data/
SUMMARY:PAC-Bayes generalization inequalities with data-dependent priors\,
and effect size quantification for interrupted time series analysis with a
pplication to COVID-19 data [ \n Graduate Student Seminar\n Semina
rs\n \n ]
DESCRIPTION:By: PhD Yael Travis-Lumer\n Advisors: Prof. Yair Goldberg\n Whe
re: Conference room at the statistics laboratory\, Cooper building\, entra
nce floor From:\nTechnion\nAbstract:\n\nDuring the first part of my PhD\,
I studied statistical inference for machine learning algorithms from a the
oretical point of view: (i) quantifying the uncertainty of machine learnin
g algorithms\, such as kernel machines\, using Bayesian Statistics and Gau
ssian processes\, (ii) estimating hyper-parameters using the Empirical Bay
es approach\, and (iii) developing novel generalization bounds using PAC-B
ayes theory.\n\nLater on\, I shifted to studying the effects of the COVID-
19 pandemic\, and its associated restrictions\, on different public health
outcomes. We used an interrupted time series (ITS) analysis which is a ti
me series regression model that aims to evaluate the effect of an interven
tion on an outcome of interest. Additionally\, we developed a methodology
to quantify the effect size in ITS. This effect size is derived from the I
TS model-based fitted values\, and the predicted counterfactual values (th
e expected values had the pandemic not occurred). Finally\, we applied our
method to national data to quantify the effect size of the COVID-19 perio
d on several public health outcomes including suicide attempts\, schizophr
enia\, antidepressants\, spontaneous abortions\, and mortality.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:258@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20220912T113000
DTEND;TZID=Asia/Jerusalem:20220912T121500
DTSTAMP:20220904T060035Z
URL:https://dds.technion.ac.il/iemevents/kernel-machines-with-missing-data
/
SUMMARY:Kernel machines with missing data [ \n Graduate Student Seminar
\n Seminars\n \n ]
DESCRIPTION:By: PhD Tiantian Liu\n Advisors: Prof. Yair Goldberg\n Where: C
onference room at the statistics laboratory\, Cooper building\, entrance f
loor From:\nTechnion\nAbstract:\n\nMissing data arises in many situations
and poses challenges in data analysis. It may seriously compromise inferen
ces if not handled appropriately. Kernel machines\, which are best known b
y the support vector machines\, have appearing advantages\, such as comput
ational ease and robustness with respect to distributional assumption. In
this research\, we develop new kernel machines to solve inferential proble
ms under three different types of missing data.\n\nThe first type of missi
ng data concerns with missing responses. We develop two new kernel machine
s\, which can be used for both regression and classification. The first pr
oposed kernel machine uses only the complete cases. It is subject to some
assumption limitations. The second proposed one is a doubly-robust kernel
machine which overcomes such limitations regardless of the misspecificatio
n of either the missing mechanism or the conditional distribution of the r
esponse. The second type of missing data considers the occurrence of missi
ng data in covariates. We develop a family of doubly robust kernel machine
s for classification assuming that the missing mechanism is missing at ran
dom. We construct a novel convex augmented loss function using inverse pro
bability weighting\, multiple imputation\, and surrogacy. The third type o
f missing data concerns a special case of missing responses in multiple in
stance learning\, where only one summarized response of a group (bag) is o
bserved. We cast the multiple instance problem as a classification with no
nignorable missing responses problem and develop three versions of the EM
algorithm using linear\, kernel machine\, and neural network classifiers t
o accommodate different levels of the data complexity.\n\n \;\n\n
\;\n\n \;
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:266@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20221109T133000
DTEND;TZID=Asia/Jerusalem:20221109T143000
DTSTAMP:20221103T063522Z
URL:https://dds.technion.ac.il/iemevents/semantic-enhancement-of-vision-tr
ansformers-relevance-maps-improves-interpretability-and-classification-cap
abilities/
SUMMARY:Semantic Enhancement of Vision Transformers Relevance Maps Improves
Interpretability and Classification Capabilities [ \n Graduate Studen
t Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Igor Drozdov\n Advisors: Prof. Tamir Hazan\n Where:
Bloomfield 152 From:\nTechnion\nAbstract:\n\nThese days visual classificat
ion models often tend to overfit to a certain degree. This may lead to a c
oncerning phenomenon\, where models may achieve high accuracy in classific
ation tasks despite focusing on non-relevant features of the input with re
spect to the prediction. In this work\, we propose a way to improve interp
retability and classification ability of these models in a unsupervised ma
nner by making the models to focus on the semantically meaningful regions
of the input with respect to the prediction. It is done as finetuning step
\, which doesn't require any additional annotated labels or masks. Specifi
cally\, our work (i) leverages the semantic knowledge captured in CLIP to
encourage the relevancy maps to rely on the semantically related foregroun
d regions of the input (ii) shows we can increase classification performan
ce once we improve the relevancy maps (iii) proposes novel approach to imp
rove relevancy maps without any additional data or training via graph bipa
rtition.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:270@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20221110T100000
DTEND;TZID=Asia/Jerusalem:20221110T110000
DTSTAMP:20221106T082124Z
URL:https://dds.technion.ac.il/iemevents/cebab-estimating-the-causal-effec
ts-of-real-world-concepts-on-nlp-model-behavior/
SUMMARY:CEBaB: Estimating the Causal Effects of Real-World Concepts on NLP
Model Behavior [ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc Eldar Abraham\n Advisors: Prof. Roi Reichart\n Where:
Bloomfield 526 From:\nTechnion\nAbstract: The increasing size and complexi
ty of modern ML systems has improved their predictive capabilities but mad
e their behavior harder to explain. Many techniques for model explanation
have been developed in response\, but we lack clear criteria for assessing
these techniques. In this paper\, we cast model explanation as the causal
inference problem of estimating causal effects of real-world concepts on
the output behavior of ML models given actual input data. We introduce CEB
aB\, a new benchmark dataset for assessing concept-based explanation metho
ds in Natural Language Processing (NLP). CEBaB consists of short restauran
t reviews with human-generated counterfactual reviews in which an aspect (
food\, noise\, ambiance\, service) of the dining experience was modified.
Original and counterfactual reviews are annotated with multiply-validated
sentiment ratings at the aspect-level and review-level. The rich structure
of CEBaB allows us to go beyond input features to study the effects of ab
stract\, real-world concepts on model behavior. We use CEBaB to compare th
e quality of a range of concept-based explanation methods covering differe
nt assumptions and conceptions of the problem\, and we seek to establish n
atural metrics for comparative assessments of these methods.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:267@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20221113T130000
DTEND;TZID=Asia/Jerusalem:20221113T133000
DTSTAMP:20221106T081954Z
URL:https://dds.technion.ac.il/iemevents/adverse-reactions-can-we-find-the
-effects-from-social-media/
SUMMARY:Assessing causal effects of medical interventions from social media
: The contribution of different algorithmic approaches [ \n Graduate S
tudent Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Yael Kiselman\n Advisors: Dr. Elad Yom-Tov\n Where:
Bloomfield 527 From:\nTechnion\nAbstract: The gold-standard approach to es
tablishing causality is through a randomized controlled trial (RCT) where
participants are randomly assigned to either a treatment or a control grou
p. Unfortunately\, in many cases only observational data (e.g.\, social me
dia) is available\, but these data include potential sources of bias. Seve
ral methods have been proposed to mitigate the effect of such biases\, for
example\, the Self Controlled Case Series method. In our work we compare
three approaches for causal discovery and apply them to social media data
(specifically\, Reddit) to identify side effects of medical interventions.
Our results show that all examined methods replicate known findings from
the literature\, except for cases where sample sizes are insufficient. Thu
s\, our work demonstrates the usefulness of social media data\, analyzed b
y appropriate algorithmic approaches\, to facilitate the discovery of adve
rse reactions of medical interventions through real-world data.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:268@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20221127T123000
DTEND;TZID=Asia/Jerusalem:20221127T130000
DTSTAMP:20221106T081726Z
URL:https://dds.technion.ac.il/iemevents/sparking-creativity-encouraging-c
reative-idea-generation-through-automatically-generated-word-recommendatio
ns/
SUMMARY:Sparking Creativity: Encouraging Creative Idea Generation through A
utomatically Generated Word Recommendations [ \n Graduate Student Semi
nar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Talia Wise\n Advisors: Prof. Yoed Kenett\n Where: Bl
oomfield 527 From:\nTechnion\nAbstract:\n\nCreative block is a familiar fo
e to any who attempt to create\, and is especially related to “writers b
lock”. While significant effort has been focused on developing methods t
o break such blocks\, it remains an active challenge. The current study pr
esents a proof-of-concept for a cognitive network sciences-based online al
gorithm that aims to spark creative ideas: Once a participant “runs out
” of ideas in a creative idea generation task\, our algorithm suggests w
ord-recommendations to prime new ideas. These word-recommendations are eit
her towards or away from previous ideas\, as well as close or far from the
target object\, based on a conceptual space extracted from the participan
ts responses using online text analysis. Our results indicate that the loc
ation of word-recommendations affects the fluency and creativity of ones
’ ideas. In addition\, we show how participants fluent intelligent and c
reativity levels affect how they many benefit from these word-recommendati
ons.\n\n \;\n\n \;
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:272@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20221127T130000
DTEND;TZID=Asia/Jerusalem:20221127T133000
DTSTAMP:20221108T071922Z
URL:https://dds.technion.ac.il/iemevents/development-of-a-machine-learning
-model-to-assess-crohns-disease-severity-from-magnetic-resonance-enterogra
phy-data/
SUMMARY:Development of a machine-learning model to assess Crohn's disease s
everity from Magnetic Resonance Enterography data [ \n Graduate Studen
t Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Itai Guez\n Advisors: Prof. Tamir Hazan and Prof. Mot
i Freiman\n Where: Bloomfield 527 From:\nTechnion\nAbstract:\nRecurrent at
tentive non-invasive observation of intestinal inflammation is essential f
or the proper management of Crohn's disease (CD).\n\nThe goal of this stud
y was to develop and evaluate a multi-modal machine-learning (ML) model to
assess ileal CD endoscopic activity by integrating information from Magne
tic Resonance Enterography (MRE) and biochemical biomarkers.\n\nWe obtaine
d MRE\, biochemical and ileocolonoscopy data from the multi-center ImageKi
ds study database.\n\nWe developed an optimized multimodal fusion ML model
to non-invasively assess terminal ileum (TI) endoscopic disease activity
in CD from MRE data. We determined the most informative features for model
development using a permutation feature importance technique.\n\nThe opti
mized fusion model performed better than the clinically recommended model
determined by both a better median test MSE distribution and a better aggr
egated AUC over the folds.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:273@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20221208T101500
DTEND;TZID=Asia/Jerusalem:20221208T111500
DTSTAMP:20221201T115433Z
URL:https://dds.technion.ac.il/iemevents/hyper-networks-for-out-of-distrib
ution-generalization-in-natural-language-processing/
SUMMARY:Hyper Networks for Out of Distribution Generalization in Natural La
nguage Processing [ \n Graduate Student Seminar\n Seminars\n \n
]
DESCRIPTION:By: M.Sc. Tomer Volk\n Advisors: Roi Reichart\n Where: Bloomfie
ld 526 From:\nTechnion\nAbstract:\n\nWhile Natural Language Processing (NL
P) algorithms keep reaching unprecedented milestones\, out-of-distribution
generalization is still challenging. In this talk we discuss the problem
of multi-source adaptation to unknown domains: Given labeled data from mul
tiple source domains\, we aim to generalize to data drawn from target doma
ins that are unknown to the algorithm at training time. We present an algo
rithmic framework based on example-based Hypernetwork adaptation: Given an
input example\, a T5 encoder-decoder first generates a unique signature w
hich embeds this example in the semantic space of the source domains\, and
this signature is then fed into a Hypernetwork which generates the weight
s of the task classifier. In an advanced version of our model\, the learne
d signature also serves for improving the representation of the input exam
ple. In experiments with two tasks\, sentiment classification and natural
language inference\, across 29 adaptation settings\, our algorithms substa
ntially outperform existing algorithms for this adaptation setup.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:287@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20221213T130000
DTEND;TZID=Asia/Jerusalem:20221213T140000
DTSTAMP:20221204T080000Z
URL:https://dds.technion.ac.il/iemevents/turn-it-onn-ordered-neural-networ
ks-for-simple-and-efficient-model-scaling/
SUMMARY:Turn it ONN: Ordered Neural Networks for Simple and Efficient Model
Scaling [ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc Hadar Sinai\n Advisors: Tamir Hazan\n Where: Behaviora
l Economics Labs\, Lady Davis Building\, Meeting Room From:\nTechnion\nNeu
ral networks usually have an inherent tradeoff between computational cost
and performance. Therefore\, it has become common to offer multiple models
derived from the same architecture\, where each pretrained model offers a
different tradeoff between the two. This process is very tedious as each
checkpoint must be separately trained. Therefore only a few checkpoints ar
e usually released and one cannot easily find a model that best fits his c
omputing resources.\nIt has been previously shown that by using Nested Dro
pout\, one can learn an ordered representation that can be adaptively comp
ressed while retaining its performance.\nIn this work\, we propose Ordered
Neural Networks (ONN)\, a neural network trained with nested dropout appl
ied on all learnable layers.\nUsing ONN\, one can train a single large che
ckpoint and get a full set of possible checkpoints. We show the effectiven
ess of this method on several benchmarks and further analyze the learned r
epresentation
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:295@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20221214T133000
DTEND;TZID=Asia/Jerusalem:20221214T143000
DTSTAMP:20221207T071426Z
URL:https://dds.technion.ac.il/iemevents/coordinating-autonomous-robots-wi
th-social-laws/
SUMMARY:Coordinating Autonomous Robots with Social Laws [ \n Graduate S
tudent Seminar\n Seminars\n \n ]
DESCRIPTION:By: Ph.D. Ronen Nir\n Advisors: Erez Karpas\n Where: Cognitive
Robotics Lab\, Cooper building From:\nTechnion\nAbstract:\nRobots may inte
rfere with each other if they act within a shared environment without prop
er coordination. Enacting Social Laws that regulate robot behavior is one
way to achieve coordination. A social law is robust if it prevents all pos
sible conflicts. As a result\, a robust social law lets the robots plan wi
thout considering the other robots' actions in the system. That makes mult
i-robot planning much more manageable. Previous work on social law verific
ation examined only the case of boolean state variables and instantaneous
actions. However\, many real-world problems require reasoning with numeric
variables and actions with durations.\nThis work aims to promote social l
aws in two directions. First\, we proposed robustness verification methods
that consider more realistic settings. Second\, we demonstrate new ways f
or social law synthesis\, i.e.\, find a robust social law for a given mult
i-robot system.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:292@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20221214T143000
DTEND;TZID=Asia/Jerusalem:20221214T153000
DTSTAMP:20221201T115409Z
URL:https://dds.technion.ac.il/iemevents/shrinkage-towards-eigenspaces-in-
the-high-dimensional-linear-model/
SUMMARY:Shrinkage towards eigenspaces in the high dimensional linear model
[ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Elad Davidson\n Advisors: Elad Davidson\n Where: Bloo
mfield 526 From:\nTechnion\n
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:283@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20221214T143000
DTEND;TZID=Asia/Jerusalem:20221214T153000
DTSTAMP:20221201T115416Z
URL:https://dds.technion.ac.il/iemevents/on-theoretical-foundations-of-soc
ial-laws/
SUMMARY:On Theoretical Foundations of Social Laws [ \n Graduate Student
Seminar\n Seminars\n \n ]
DESCRIPTION:By: Ph.D. Alexander Tuisov\n Advisors: Erez Karpas\n Where: Cog
nitive Robotics Lab (Cooper building) From:\nTechnion\nAbstract: \nMultipl
e agents operating in a shared environment can interfere with each other's
ability to reach their goals. One of the approaches to address this issue
is enacting a social law -- a set of rules that restricts some possible b
ehaviors of the agents. A social law that ensures that each agent can achi
eve its goal\, regardless of what the other agents do\, is called robust.\
nIn this work\, we strive to uncover deeper theoretical foundations of soc
ial laws in the context of STRIPS-like environments. We start by pointing
out and correcting a mistake in the previous work. Furthermore\, we establ
ish the semantics behind the waiting mechanism and connect it to the agent
's sensing capabilities. Using these insights\, we also derive methods for
faster robustness validation. Finally\, we reason about the social law ro
bustness in a reactive environment\, getting one step closer to real-world
scenarios.\n
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:288@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20221218T123000
DTEND;TZID=Asia/Jerusalem:20221218T130000
DTSTAMP:20221207T075929Z
URL:https://dds.technion.ac.il/iemevents/cognitive-limitations-vs-cognitiv
e-sophistication-on-the-relationship-between-memory-and-maximization-in-de
cision-making-tasks/
SUMMARY:Cognitive limitations vs. cognitive sophistication: On the relation
ship between memory and maximization in decision making tasks [ \n Gra
duate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc Nitzan Strauss\n Advisors: Kinneret Teodorescu and Or
i Plonsky\n Where: Bloomfield 527 From:\nTechnion\nAbstract:\n\nPast resea
rch on repeated decisions finds two robust deviations from maximization (i
.e.\, probability matching and underweighting of rare events). People’s
decisions rely only on a small subset of their past experiences\, rather t
han on all experiences\, can account for these findings. Why is that so? T
he common explanation is due to cognitive limitations\, such as limited me
mory capacity\, forgotten earlier experiences. However\, recent studies su
ggest an alternative approach\, the cognitive sophistication approach\, ac
cording to which a smart and complex strategy of seeking for patterns unde
rlies the tendency to rely on small samples. Whereas the cognitive limitat
ions approach predicts that people with greater cognitive abilities exhibi
t less deviations from maximization\, while the cognitive sophistication a
pproach predicts the opposite. This exploratory study examined the relatio
n between deviations from maximization and cognitive abilities. Some of th
e findings provide support for the cognitive limitations approach\, while
others support the cognitive sophistication approach. The current mixed fi
ndings emphasis the need for further research.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:280@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20221218T130000
DTEND;TZID=Asia/Jerusalem:20221218T133000
DTSTAMP:20221201T115335Z
URL:https://dds.technion.ac.il/iemevents/stock-movement-price-prediction-w
ith-behavioral-features/
SUMMARY:Stock Movement Price Prediction with behavioral features [ \n G
raduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Rania Khoury\n Advisors: Ori Plonsky and Margarita Os
adchy\n Where: Bloomfield 527 From:\nTechnion\nAbstract: \nStock market pr
ediction has been a challenging problem that has caught the attention of m
any researchers. Building an accurate predictive model would contribute gr
eatly to the economic stability of the countries. Technical Analysis\, Lin
ear models\, Machine Learning are just a few examples of the different app
roaches that have been tried to predict the stock market. Some researchers
introduced into their models features based on sentiment analysis of soci
al media tweets\, proving that accounting for investors' behavior can help
predictive models. In addition to investors' sentiment\, it is possible t
hat incorporating theory-driven behavioral insights into predictive models
may also improve performance.\nIn this work we show the effect of incorpo
rating behavioral features to an Attentive LSTM network for stock predicti
on.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:296@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20221218T134500
DTEND;TZID=Asia/Jerusalem:20221218T141500
DTSTAMP:20221211T063842Z
URL:https://dds.technion.ac.il/iemevents/refining-binomial-confidence-inte
rvals/
SUMMARY:Refining binomial confidence intervals [ \n Graduate Student Se
minar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Almog Peer\n Advisors: David Azriel\n Where: Bloomfie
ld 527 From:\nTechnion\nAbstract:\n\nEstimation of the probability of the
binomial distribution is a basic problem\, which appears in almost all int
roductory statistics course\, and is preformed frequently in various studi
es. In some cases\, the parameter of interest is a difference between two
probabilities\, and the current work studies the construction of confidenc
e intervals for this parameter when the sample size is small. Our goal is
to find the shortest confidence intervals under the constraint of coverage
probability being large than a predetermined level. For the two-sample ca
se there is no known algorithm that achieves this goal\, but rather differ
ent heuristics procedures were suggested.\n\nThis work fills this gap\, na
mely\, we construct an algorithm that computes for small sample-sizes the
optimal confidence intervals. These are compared to the sub-optimal proced
ures that were suggested previously\, and we find that the improvement is
generally not very large.\n\n \;\n\n \;\n\n \;
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:282@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20221225T130000
DTEND;TZID=Asia/Jerusalem:20221225T133000
DTSTAMP:20221201T115320Z
URL:https://dds.technion.ac.il/iemevents/pose-estimation-for-surgical-trai
ning/
SUMMARY:Pose Estimation For Surgical Training [ \n Graduate Student Sem
inar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Eddie Bkheet\n Advisors: Shlomi Laufer\n Where: Bloom
field 527 From:\nTechnion\nAbstract:\nPurpose: This research aims to utili
ze computer vision algorithms for the automated training of surgeons and t
he analysis of surgical footage.\n\nMethods: Using pre-trained models we c
reate our own dataset of 100 open surgery simulation videos with 2D hand p
oses. We also assess the ability of pose estimations to segment surgical v
ideos into gestures and tool-usage segments and compare them to sensors an
d I3D features. Furthermore\, we introduce novel surgical skill proxies st
emming from domain experts’ training advice.\n\nResults: State-of-the-ar
t gesture segmentation accuracy on the Open Surgery Simulation dataset. Th
e introduced surgical skill proxies presented significant differences for
novices compared to experts and produced actionable feedback for improveme
nt.\n\nConclusion: Pose estimations achieved comparable results to physica
l sensors while being remote and markerless. Surgical skill proxies that r
ely on hand poses proved they can be used to work towards automated traini
ng feedback.\n\n+ Google Calendar + ICal Export
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:301@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20221229T153000
DTEND;TZID=Asia/Jerusalem:20221229T163000
DTSTAMP:20221215T111747Z
URL:https://dds.technion.ac.il/iemevents/how-reliable-is-the-endocast-for-
inferring-brain-anatomy-a-deep-learning-approach/
SUMMARY:How reliable is the endocast for inferring brain anatomy? a deep le
arning approach [ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Itay Hadar\n Advisors: Assaf Marom \n Where: Anatomy
Education and Research Center (RMC) From:\nTechnion\nAbstract:\n\nIn our p
roject\, we design a full pipeline to predict a human brain anatomy from t
he human skull by using a deep convolutional network that encodes informat
ion from the skull and decodes that information to a brain image. The Impo
rtance of the suggested study is twofold. First\, we will be able to confi
rm whether the endocast is indeed a reliable proxy for the human brain\, a
n assumption that has yet to be validated quantitatively. Second\, if our
results confirm the endocast as a reliable source for neuroanatomical data
\, a new set of question will open\, as our algorithm could possibly be ge
neralized and used for inferring fossil species' brains from fossilized en
docast.\n\n \;\n\n \;
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:293@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230108T123000
DTEND;TZID=Asia/Jerusalem:20230108T130000
DTSTAMP:20221207T071132Z
URL:https://dds.technion.ac.il/iemevents/self-supervised-future-generator-
for-human-action-segmentation/
SUMMARY:Self-supervised future generator for human action segmentation [ \n
Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Or Berman\n Advisors: Shlomi Laufer\n Where: Bloomfie
ld 527 From:\nTechnion\nAbstract:\n\nThe ability to locate and classify ac
tion segments in long untrimmed video is of particular interest to many ap
plications such as autonomous cars\, robotics and healthcare applications.
Today\, the most popular pipeline for action segmentation is composed of
encoding the frames into feature vectors\, which are then processed by a t
emporal model for segmentation. In this paper we present a self-supervised
method that comes in the middle of the standard pipeline and generated re
fined representations of the original feature vectors. Experiments show th
at this method improves the performance of existing models on different su
b-tasks of action segmentation\, even without additional hyper parameter t
uning.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:291@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230108T130000
DTEND;TZID=Asia/Jerusalem:20230108T133000
DTSTAMP:20221207T070923Z
URL:https://dds.technion.ac.il/iemevents/automatic-evaluation-of-laparosco
pic-skills-in-the-task-of-peg-transfer/
SUMMARY:Automatic Evaluation of Laparoscopic Skills in the Task of Peg Tran
sfer [ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Aviad Lazar\n Advisors: Shlomi Laufer\n Where: Bloomf
ield 527 From:\nTechnion\nAbstract:\nFundamentals of Laparoscopic Surgery
(FLS) box trainer is a well-accepted method for training and evaluating la
paroscopic skills. It mandates an observer that will measure and evaluate
the trainee's performance. This study aimed to assess whether computer vis
ion (CV) and artificial intelligence (AI) may be used to automatically mea
sure performance in the FLS box trainer\, specifically for Peg Transfer. M
easuring performance in the Peg Transfer task includes time and a penalty
for dropping objects. Four groups of metrics were defined and measured aut
omatically using CV: Time\, Grasper Movement Speed\, Path Efficiency\, and
Grasper Coordination. Validity was assessed by dividing participants to 3
groups of experience levels\, and comparing performance between groups w.
r.t the defined metrics.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:284@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230115T123000
DTEND;TZID=Asia/Jerusalem:20230115T130000
DTSTAMP:20221229T100509Z
URL:https://dds.technion.ac.il/iemevents/fusion-from-a-condorcet-perspecti
ve/
SUMMARY:Fusion from a Condorcet Perspective [ \n Graduate Student Semin
ar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Liron Tyomkin\n Advisors: Oren Kurland \n Where: Bloo
mfield 527 From:\nTechnion\nAbstract:\n\nThe fusion task is to merge docum
ent lists retrieved from a corpus in response to a query. Despite the few
decades of research work on fusion\, there are still no theoretical tools
to formally analyze fusion methods. We use the Condorcet voting rule as a
first step to this end. We then present an in-depth empirical analysis of
TREC runs and existing fusion methods with respect to Condorcet. Additiona
l exploration shows that a simple approach to turn existing fusion methods
to Condorcet methods yields highly effective performance.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:285@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230115T130000
DTEND;TZID=Asia/Jerusalem:20230115T133000
DTSTAMP:20221226T071800Z
URL:https://dds.technion.ac.il/iemevents/cluster-based-document-retrieval/
SUMMARY:Cluster-Based Document Retrieval [ \n Graduate Student Seminar\
n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Egor Markovskiy\n Advisors: Shoham Sabach and Oren Ku
rland\n Where: Bloomfield 527 From:\nTechnion\nAbstract: \nThe common appr
oach of using clusters of similar documents for ad hoc document retrieval
is to rank the clusters in response to the query\; then\, the cluster rank
ing is transformed to document ranking. We present a novel supervised appr
oach to transform cluster ranking to document ranking. The approach allows
to simultaneously utilize different clusterings and the resultant cluster
rankings\; this helps to improve the modeling of the document similarity
space. Empirical evaluation shows that using our approach results in perfo
rmance that substantially transcends the state-of-the-art in cluster-based
document retrieval.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:310@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230117T113000
DTEND;TZID=Asia/Jerusalem:20230117T123000
DTSTAMP:20230102T160448Z
URL:https://dds.technion.ac.il/iemevents/examining-the-feasibility-of-mach
ine-learning-classifiers-in-typical-and-clinical-cognition/
SUMMARY:Examining the Feasibility of Machine Learning Classifiers in Typica
l and Clinical Cognition [ \n Graduate Student Seminar\n Seminars\
n \n ]
DESCRIPTION:By: M.Sc. Alon Tsaizel\n Advisors: Yoed Kenett\n Where: Cogniti
ve Complexity Lab\, Lady Davis\, 0 Floor From:\nTechnion\nAbstract:\n\nMac
hine learning methods are utilized in a wide variety of fields\, such as h
ealthcare\, finance\, and marketing\, and have shown great success in them
. They allow automating analytical model construction in order to draw qua
ntitative conclusions from data. In this research we examine the performan
ce of various machine learning models\, and specifically classifiers\, on
cognitive data\, in both typical and clinical populations.\n\nIn Study 1\,
focusing on typical (healthy) population\, we use machine learning models
to predict one’s level of intelligence\, specifically fluid intelligenc
e\, based on various cognitive and behavioral measures\, such as personali
ty and behavior. We find that intelligence can be predicted quite well usi
ng only a small subset of features.\n\nIn Study 2\, focusing on clinical p
opulations\, we use machine learning models in conjuncture with a multidim
ensional representation of the mental lexicon (via a cognitive multiplex n
etwork model) to predict whether a patient has dementia and if so\, the sp
ecific type of dementia they have\, based on a simple semantic fluency tas
k. We find that these models can predict almost perfectly whether a patien
t has dementia\, and considerably well the specific type of dementia.\n\nO
verall\, in both studies we demonstrate the significance of applying machi
ne learning algorithms to study cognition\, in both typical and clinical p
opulations.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:314@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230119T102000
DTEND;TZID=Asia/Jerusalem:20230119T112000
DTSTAMP:20230105T121349Z
URL:https://dds.technion.ac.il/iemevents/state-of-affairs-in-domain-adapta
tion-for-nlp/
SUMMARY:State of Affairs in Domain Adaptation for NLP [ \n Graduate Stu
dent Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Naveh Porat\n Advisors: Roi Reichart\n Where: Bloomfi
eld 526 From:\nTechnion\nAbstract: Recent advances in Natural Language Pro
cessing (NLP) have brought significant improvements in the performance of
models on many NLP tasks. The improvements aren’t limited just to higher
scores\, but also in the models’ ability to generalize based on fewer s
amples and their robustness to many types of fluctuations in data and trai
ning. This raises the question: Are these models robust to domain shifts?
That is\, when trained on data from one domain\, can they still maintain t
he same performance on data from other domains?\n\nWe propose a thorough e
xamination of the current state of the Domain Adaptation problem in NLP\,
as well as a varied suite of new evaluation tasks to promote future resear
ch on the subject.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:290@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230122T123000
DTEND;TZID=Asia/Jerusalem:20230122T130000
DTSTAMP:20221201T115220Z
URL:https://dds.technion.ac.il/iemevents/clinical-prediction-using-transfo
rmer-based-survival-analysis/
SUMMARY:Clinical prediction using transformer-based survival analysis [ \n
Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Moshe Zisser\n Advisors: Dvir Aran\n Where: Bloomfiel
d 527 From:\nTechnion\nAbstract:\nHealthcare providers are increasingly us
ing machine learning to predict patient outcomes and provide meaningful in
terventions. Deep learning may improve prediction performance by learning
representations of longitudinal health records capturing patients' current
health status and future risks.\n\nNatural language processing tasks are
well performed by deep learning models such as the transformer. Contrary t
o this\, transformers are rarely used in healthcare.\n\nWe introduce STRAF
E\, or Survival Analysis Transformer Based Architecture For EHR\, which ut
ilizes the transformer computational power to perform time-to-event predic
tions of patient deterioration in Chronic Kidney Disease.\n\nBy analyzing
EHR data end-to-end\, this study will contribute to the prediction of medi
cal events without performing feature engineering. Additionally\, the arch
itecture was developed based on OMOP-formatted data to enable its applicat
ion to other datasets with the same format.\n\n \;
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:311@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230122T130000
DTEND;TZID=Asia/Jerusalem:20230122T133000
DTSTAMP:20230102T160626Z
URL:https://dds.technion.ac.il/iemevents/price-of-anarchy-in-si-model/
SUMMARY:Price of anarchy in SI model [ \n Graduate Student Seminar\n
Seminars\n \n ]
DESCRIPTION:By: M.Sc. Daniel Ablin\n Advisors: Rann Smorodinsky\n Where: Bl
oomfield 527 From:\nTechnion\nAbstract:\n\nDuring the current COVID-19 pan
demic the most prevalent means of reaction around the world was to impose
social distancing. This has been done by means of rules requiring the use
of masks and up to a curfew.\n\nThis top-down approach\, where the governm
ent dictates the behavior of individuals can be contrasted with a natural
alternative\, a grass roots approach\, whereby the responsibility for soci
al distancing is given to the individuals. Whereas the risk associated wit
h a grass roots approach is quite obvious (anarchic behavior) the potentia
l benefits are high. Mainly\, in such an approach each individual can adop
t the level and means of social distancing he takes and adjust it to the r
isk he bears and to the alternative cost.\n\nThe fact that the latter alte
rnative\, where self-distancing is voluntary\, was hardly chosen suggests
that in most countries\, the underlying assumption is that it would not wo
rk as well. In this thesis we plan to investigate this underlying assumpti
on.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:312@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230201T093000
DTEND;TZID=Asia/Jerusalem:20230201T103000
DTSTAMP:20230102T160907Z
URL:https://dds.technion.ac.il/iemevents/optimization-methods-for-solving-
the-nonsmooth-nonconvex-sensor-localization-problem-2/
SUMMARY:Optimization Methods for Solving the Nonsmooth Nonconvex Sensor Loc
alization Problem [ \n Graduate Student Seminar\n Seminars\n \n
]
DESCRIPTION:By: M.Sc. Omer Cicelsky\n Advisors: Shoham Sabach\n Where: ZOOM
From:\nTechnion\nAbstract:\n\nThe sensor localization problem is the prob
lem of finding the best location estimation of radiating sensors\, from se
veral noisy range measurements of distances collected using a network. Eac
h sensor has a power supply and a radio transceiver that enable it to moni
tor its immediate vicinity. In this project\, we consider two types of pro
blems: the first one refers to the problem of finding the best estimate fo
r the location of only one radiating source this type of problem is known
in the literature as Single Source Localization (SSL) problem. For such pr
oblems\, we assume that the location of the anchors are known exactly. In
the second type\, we deal with networks with a large number of non-anchor
sensors\, this type of problem is referred to as Wireless Sensor Network L
ocalization (WSNL) problem. In this work we tackled both cases by designin
g novel range-based algorithms that utilize Time-of-Arrival information\,
in order to determine the position of each of the sensors.\n\n \;\n\nZ
oom Link\n\nhttps://technion.zoom.us/j/96810699297
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:318@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230201T143000
DTEND;TZID=Asia/Jerusalem:20230201T153000
DTSTAMP:20230115T061228Z
URL:https://dds.technion.ac.il/iemevents/what-can-we-learn-by-analyzing-in
ternal-organizational-communication-a-data-science-analysis-of-slack/
SUMMARY:What Can We Learn by Analyzing Internal Organizational Communicatio
n? A Data Science Analysis of Slack [ \n Graduate Student Seminar\n
Seminars\n \n ]
DESCRIPTION:By: M.Sc. Rani Khoury\n Advisors: Anat Rafaeli and Ofra Amir\n
Where: Bloomfield 527 From:\nTechnion\nAbstract:\n\nWhat information does
the internal communication between employees of an organization convey? Ho
w do these develop and evolve over time as the company grows and formalize
s? We study these questions through an in-depth and data-science analysis
of four years’ worth of internal Slack public channels communication in
a mid-size high-tech firm. We empirically examine four core aspects of emp
loyees: Behavior\, Text Focus\, Text Affect\, and Communication networks.
We identify two cardinal and complimentary types of behavior (activity and
interactivity) using factor analysis and three types of focus (Internal T
asks\, External Tasks and People) using topic modelling. We find that memb
ers of different divisions maintain different norms regarding their behavi
or and regarding topics discussed. We extracted the expressed affect in em
ployee messages using sentiment analysis and analyzed their change over ti
me as the firm evolves and formalizes. We then use network analysis tools
to identify patterns of exchanges between users and emergent sub-communiti
es within the firm. We find that the growth and maturity of the firm leads
to an emergent pattern of an increasing tendency among employees to inter
act within their own division rather than other divisions and a generation
of sub-communities with respect to the divisional structure of the firm.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:323@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230212T133000
DTEND;TZID=Asia/Jerusalem:20230212T143000
DTSTAMP:20230201T105247Z
URL:https://dds.technion.ac.il/iemevents/automatic-performance-evaluation-
of-the-fls-simulator-using-computer-vision-and-strain-analysis/
SUMMARY:Automatic Performance Evaluation of The FLS Simulator Using Compute
r Vision And Strain Analysis [ \n Graduate Student Seminar\n Semin
ars\n \n ]
DESCRIPTION:By: M.Sc. Liran Halperin \n Advisors: Shlomi Laufer \n Where:
Scalpel lab\, Lady Davis\, 0 floor\, room 105 From:\nTechnion\nAbstract:
\n\nThe Fundamentals of Laparoscopic Surgery (FLS) box trainer is a standa
rdized simulator for training and assessing basic laparoscopic skills. Thi
s work is comprised from two studies that provide new ways to assess the p
erformance of the practitioner using computer vision and statistical analy
sis.\n\nThe aim of the first study is to provide automated feedback on the
intracorporeal suture exercise in the FLS simulator using deep learning a
lgorithms. Different metrics were developed with the goal of providing inf
ormative feedback to the user. The automation of the feedback will allow s
tudents to practice at any time without the supervision of experts. Obje
ct detection and semantic segmentation were used to collect statistics on
the practitioner's performance. Three task specific metrics were defined.
Good agreement between the human labeling and the different algorithms was
achieved.\n\nThe aim of the second study is to develop a system that meas
ure how much force is applied on the graspers while performing the peg tra
nsfer exercise in the FLS simulator. We installed strain gages on the gras
per that allows us to measure the forces the practitioner applies on the g
rasper. We measured the strains of residents and senior surgeons performin
g the exercise. Statistical analysis was performed on the results. The ana
lysis shows significant differences of the way forces are applied on diffe
rent materials\, during transferring the triangles of the exercise from on
e hand to another and after repeating the exercise.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:325@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230316T133000
DTEND;TZID=Asia/Jerusalem:20230316T140000
DTSTAMP:20230305T121515Z
URL:https://dds.technion.ac.il/iemevents/conversational-retrieval/
SUMMARY:Conversational Retrieval [ \n Graduate Student Seminar\n Se
minars\n \n ]
DESCRIPTION:By: M.Sc. Tomer Gureivch \n Advisors: Oren Kurland\n Where: ZOO
M From:\nTechnion\nAbstract:\n\nConversational retrieval is a process wher
e a user posts a series of\nqueries/questions to a retrieval system. Our w
ork addresses two challenges unique to this setting. First\, we designed a
novel retrieval framework for conversational retrieval. We combine multip
le representations for a given query and multiple representations of the d
ocument to improve retrieval effectiveness. Second\, we addressed the nove
l\nchallenge of predicting the effectiveness of retrieval performed for\na
query in the conversation\; i.e.\, estimating retrieval effectiveness wit
h no relevance judgments. We present two prediction methods. The first uti
lizes information induced from previous queries in the conversation. The s
econd prediction method uses information induced from alternative forms of
representation for the given query.\nEmpirical evaluation attests to the
merits of our retrieval and prediction methods.\n\nZoom Link\n\nhttps://te
chnion.zoom.us/j/91808480009
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:324@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230316T140000
DTEND;TZID=Asia/Jerusalem:20230316T143000
DTSTAMP:20230305T090915Z
URL:https://dds.technion.ac.il/iemevents/predicting-query-performance-for-
cluster-based-retrieval/
SUMMARY:Predicting Query Performance For Cluster-Based Retrieval [ \n G
raduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Itay Yacov \n Advisors: Oren Kurland\, Fiana Raibe
r \n Where: ZOOM From:\nTechnion\nAbstract:\n\n \;\n\nEffectiveness e
stimation of a search performed in response to a query in the absence of
relevance judgments is a core challenge named query-performance. Two main
categories of approaches have been developed for this task: pre-retrieva
l and post-retrieval methods. Pre-retrieval methods utilize both documen
t level and corpus level statistics and aims to estimate the quality of
the search results before the search takes place. Post-retrieval methods
can also incorporate additional information induced from the actual search
results list.\n\nIn this work\, we address the prediction challenge for
cluster-based document retrieval methods in the case where the final resu
lt list is produced by utilizing information induced from clusters of simi
lar documents. We present novel prediction methods for cluster-based docum
ent retrieval which exploit the unique characteristics of the retrieval se
tting\; i.e.\, using document clusters. We demonstrate the empirical merit
s of the prediction methods we propose.\n\n \;\n\nZoom Link\n\nhttps:/
/technion.zoom.us/j/91808480009
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:332@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230328T133000
DTEND;TZID=Asia/Jerusalem:20230328T143000
DTSTAMP:20230321T130649Z
URL:https://dds.technion.ac.il/iemevents/the-impact-of-project-management-
strategy-on-project-duration/
SUMMARY:The impact of Project Management Strategy on Project Duration [ \n
Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: PhD Itai Lishner\n Advisors: Avraham Shtub \n Where: Bloomf
ield 527 From:\nTechnion\nAbstract:\n\nThe traditional model of project su
ccess assumes that there are three equally important criteria - Time\, Cos
t and Scope/Quality (the “Iron Triangle”). There are tools and techniq
ues developed to help managers satisfy these criteria and to achieve proje
ct success.\nRecent studies revealed that the traditional “Iron Triangle
” model is not always the way modern organizations measure success. In m
any organizations “Time” is the most significate factor for measuring
the success of projects.\nEstimating the duration of a project is a challe
nging task due to different sources of uncertainty\, the uniqueness of eac
h project\, and the differences between organizations’ culture and manag
ement techniques.\n\nThe study focuses on the “ripple effect” which is
characterized by an exponential increase in project duration caused by a
high volume of disruptions\, and suggests an AI based model that uses info
rmation from past projects to better estimate future project duration.\n\n
\;\n\nThis work is summarized in 3 papers:\n\nLishner\, I.\, &\; S
htub\, A. (2019). Measuring the success of Lean and Agile projects: Are co
st\, time\, scope and quality equally important?. The Journal of Modern P
roject Management\, 7(1). ISSN: 2317-3963\n\nLishner\, I.\, &\; Shtu
b\, A. (2021). The compounding effect of multiple disruptions on construct
ion projects. International Journal of Construction Management\, 1-8.\n\n
Lishner\, I.\, &\; Shtub\, A. (2022). Using an Artificial Neural Networ
k for Improving the Prediction of Project Duration. Mathematics\, 10(22)
\, 4189.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:334@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230423T123000
DTEND;TZID=Asia/Jerusalem:20230423T130000
DTSTAMP:20230416T054910Z
URL:https://dds.technion.ac.il/iemevents/dynamic-spare-parts-inventory-man
agement-utilizing-machine-health-data/
SUMMARY:Dynamic Spare Parts Inventory Management Utilizing Machine Health D
ata [ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc Avital Chernshev\n Advisors: Yale T. Herer\n Where: Bl
oomfield 527 From:\nTechnion\nAbstract:\n\nIn this research\, we develope
d ways to utilize machine health data in order to improve supply chain per
formance. This research was carried out in cooperation with an Israeli AI
company which specializes in gathering and analyzing machine health data t
o improve productivity and profitability. In many cases\, demand for spar
e parts is assumed to follow a time homogeneous Poisson process\, i.e.\, t
he time between failures is an exponentially distributed random variable w
ith a constant rate. While this has proven to be an applicable assumption
without machine health data\, it is clearly not so with this data. Almost
by definition\, the failure rate of a machine increases as its health dec
reases. Our aim was to use this now available data related to machine heal
th in order to improve spare parts inventory management policies. Clearly
\, these policies will be dynamic in relation to the machines' status. We
focus on spare parts for machines that are run with redundancy\, that is t
hat for the operation in question\, one machine runs while a (more-or-les
s) identical machine sits idle\, ready to take over if the first machine f
ails. This type of redundancy is typical in the process industry. We defin
ed a new inventory problem\, which was enabled by the company's data colle
ction and analysis capabilities. We then built a theoretical model which
allowed us to demonstrate and investigate the relationship between the par
ameters and decision variables in order to assist in the identification of
how supply chain performance can be improved. We calculated the exact st
eady-state probabilities of the Markov chain for small instances. In order
to obtain the probabilities for larger instances e.g.\, dozens of pairs o
f identical machines\, we used simulation which was validated by the exact
solution calculations. The parameters for the cost estimation were holdi
ng cost per part\, a large penalty when a pair of machines went down and a
smaller penalty for cases where the back-up machine runs but there are no
available parts for the broken machine i.e.\, temporary no-redundancy st
ate. Once the model was fixed\, we developed several decision rules which
allow the system to dynamically change the spare parts inventory levels ac
cording to the system's state. A comparison between our dynamic policy and
a static base-stock policy showed that spare parts inventory costs can b
e reduced substantially when exploiting the availability of machine health
data.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:357@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230618T123000
DTEND;TZID=Asia/Jerusalem:20230618T130000
DTSTAMP:20230618T043954Z
URL:https://dds.technion.ac.il/iemevents/covid-19-estimate-an-optimal-timi
ng-for-booster-vaccination-based-on-immunological-history/
SUMMARY:COVID-19: Estimate an optimal timing for booster vaccination based
on immunological history [ \n Graduate Student Seminar\n Seminars\
n \n ]
DESCRIPTION:By: M.Sc. Amit Berkowitz \n Advisors: Yair Goldberg\n Where: B
loomfield 526 From:\nTechnion\nAbstract:\n\nThree years into the COVID-19
pandemic\, booster vaccination policies are still derived\nmainly from the
rise in community transmission\, rather than data-driven. Particularly\,\
nvaccination policies usually do not take into account immunological histo
ry and personal characteristics\,\napart from age. In this work\, we devel
oped a survival analysis model for finding an\noptimal time for a booster
vaccination after recovery\, given personal explanatory variables. To\nfor
mulate the optimization problem for finding optimal vaccination timing we
used two hazard\nfunctions that describe the hazard of reinfection before
and after vaccination is carried out. We\nthen demonstrated the performanc
e of our model using mathematical derivations and data experiments\nunder
the Weibull model. In these experiments\, our model has shown better resul
ts\nthan standard policies both in terms of the rate of reinfection for a
given period of time and in\nterms of time when the population infection r
ate exceeds a given threshold.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:353@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230625T123000
DTEND;TZID=Asia/Jerusalem:20230625T133000
DTSTAMP:20230528T083107Z
URL:https://dds.technion.ac.il/iemevents/false-name-attacks-and-fair-alloc
ation-in-economic-mechanisms/
SUMMARY:False-name Attacks and Fair Allocation in Economic Mechanisms [ \n
Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: Ph.D. Yotam Gafni \n Advisors: Moshe Tennenholtz\, Ron Lavi
\n Where: Bloomfield 527 From:\nTechnion\nEconomic mechanism design studie
s how to design mechanisms that satisfy good societal properties\, while c
onsidering that agents may try to game the system. An ominous form of atta
ck is the 'False Name Attack'\, where agents use multiple fake identities
to interact with the system for their benefit. These attacks appear in a p
lethora of settings such as combinatorial auctions\, collaborative machine
learning protocols\, and in the formation of voting bodies. Generally\, t
rying to mitigate false-name attacks comes at the cost of significant welf
are loss. My work takes a novel approach of using uncertainty as a tool to
deter attackers\, and shows that when attackers do not know the exact det
ails of the system they attack\, they may strategically prefer to remain h
onest.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:372@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230625T133000
DTEND;TZID=Asia/Jerusalem:20230625T143000
DTSTAMP:20230619T063529Z
URL:https://dds.technion.ac.il/iemevents/are-anti-vaxxers-anti-social-on-t
he-relationship-between-immunization-against-covid-19-prosociality-and-tru
st/
SUMMARY:Are Anti-Vaxxers Anti-Social\, On The Relationship Between Immuniza
tion Against Covid-19\, Prosociality and Trust [ \n Graduate Student S
eminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Sapir gavriel\n Advisors: Amnon Maltz\, Moti Michael
i\n Where: Bloomfield 527 From:\nTechnion\nAbstract:\n\ngetting vaccinate
d against Covid-19 is effective both for self-regarding reasons (protectin
g ourselves against serious illness\, hospitalization and death) and for s
ocial reasons (protecting those around us by slowing down the spread of th
e pandemic and stopping new variants from emerging). The latter are largel
y responsible for the common view according to which the behaviour of thos
e who refuse to vaccinate is anti-social and inconsiderate. However\, it r
emains unclear whether the choice not to vaccinate is\, in fact\, predicte
d by (anti) social motives. Research on vaccine hesitancy has mostly focus
ed on self-regarding preferences\, that is\, on the perspective of individ
uals’ own wellbeing. We focus instead on other-regarding preferences (sp
ecifically altruism\, trust in others and trustworthiness) and examine exp
erimentally whether vaccine hesitancy may be predicted by such preferences
. To do so we utilize two classic games from the social-science literature
: The Dictator Game and the Trust Game. We believe that the outcomes of th
is research will shed light on understudied motives behind the decision no
t to vaccinate and assist policy makers in constructing soft intervention
tools to increase uptake rates.\n\nתקציר:\n\nבזמן מגפת הקו
רונה (COVID-19) חשיבות התחסנות הציבור באוכלו
סייה הוא קריטי בכדי להאט את התפשטות המגפ
ה. לכן מוטב לדעת מהם הגורמים אשר יכולים ל
סייע בהעלאת אחוז המתחסנים באוכלוסייה. במ
חקר זה השתמשנו בשאלון לצד משחקים מתומרצי
ם (משחק הדיקטטור\, משחק האמון) על מדגם מיי
צג של מחוסנים ולא מחוסנים בישראל\, באמצעו
ת ממשק דיגיטלי שבו השתתפו 1461 פאנליסטים מ
גיל 18 ומעלה שאת נתוניהם ניתחנו. בדקנו האם
ישנו קשר בין פרו חברתיות\, אמון בזולת\, מי
דת ראוי לאמון ואמון במשרד הבריאות לבין ה
תחסנות.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:361@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230702T123000
DTEND;TZID=Asia/Jerusalem:20230702T130000
DTSTAMP:20230607T090258Z
URL:https://dds.technion.ac.il/iemevents/modelling-social-situation-awaren
ess-in-driving-interactions/
SUMMARY:Modelling Social Situation Awareness in Driving Interactions [ \n
Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Navit Klein\n Advisors: Avi Parush \n Where: Bloomfi
eld 527 From:\nTechnion\nAbstract:\n\nThe design of self-driving vehicles
requires a comprehensive understanding of the social interactions between
drivers in resolving ambiguous encounters\, such as at unsignalized inters
ections. This research highlights the significance of social situational a
wareness (SA) as a model for understanding everyday driving interactions\,
as well as its implications on inter-driver communications\, negotiations
\, and autonomous vehicle (AV) design.\n\nUsing a dual-participant VR driv
ing simulator\, we collected data from driving encounter scenarios involve
N= 170 participants to assess their social SA during encounters that culm
inated in crossing an intersection safely.\n\nWe developed a social SA que
stionnaire to assess participants' awareness of other drivers' direction o
f approach to the intersection\, signaling\, speed and speed change\, and
heading of the vehicle.\n\nOur study results revealed statistically signif
icant relations\, reflecting the sequential perception\, and comprehension
of explicit cues\, such as signaling\, and implicit cues\, with speed cha
nge being the most substantial. These cues were highly associated with awa
reness of who entered the intersection first and expectations of the other
driver's actions. Based on these findings\, we propose a model of Social
Situation Awareness that incorporates drivers' approach\, speed\, change o
f speed\, heading\, and explicit signaling. The social SA questionnaire\,
study\, and analysis protocol developed for this work can contribute to th
e design of AVs that are situationally aware of social aspects in driving
encounters.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:362@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230702T130000
DTEND;TZID=Asia/Jerusalem:20230702T133000
DTSTAMP:20230607T094939Z
URL:https://dds.technion.ac.il/iemevents/confidence-prediction-using-mouse
-tracking-data/
SUMMARY:Confidence Prediction Using Mouse-Tracking Data [ \n Graduate S
tudent Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Lidor Asulin\n Advisors: Rakefet Ackerman\n Where: B
loomfield 527 From:\nTechnion\nAbstract: \n\nConfidence judgment is a k
ey self-reported measure in metacognitive research\, usually measured whi
le participants address a cognitive task (e.g.\, solving a set of probl
ems). When confidence cannot be explicitly elicited (e.g.\, using existin
g data of natural computerized work)\, alternative methods for confidence
prediction are needed. In this work\, we examined the extent to which mo
use-tracking data can be used to replace explicit confidence ratings in mu
ltiple-choice cognitive tasks. For this purpose\, we collected mouse-track
ing data during online problem-solving tasks and trained confidence predic
tion models using a combination of basic features (both general and task-s
pecific) and mouse-tracking features. By comparing the performance of mode
ls trained on two distinct tasks\, this research seeks to determine the ut
ility of mouse-tracking features for confidence prediction and assess the
necessity of task-specific features.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:337@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230717T183000
DTEND;TZID=Asia/Jerusalem:20230717T193000
DTSTAMP:20230615T120006Z
URL:https://dds.technion.ac.il/iemevents/first-order-methods-for-non-conve
x-optimization-problems-with-block-structure/
SUMMARY:First-Order Methods for Non-Convex Optimization Problems with Block
Structure [ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: Ph.D. Eyal Gur\n Advisors: Shoham Sabach\n Where: ZOOM From
:\nTechnion\nAbstract:\n\nThis PhD thesis focuses on first-order algorithm
s for tackling the wide class of non-convex and non-smooth optimization pr
oblems that admit a simple block structure. The thesis proposes a new gene
ral optimization scheme for tackling such problems\, called Nested Alterna
ting Minimization (NAM)\, that combines the classical Alternating Minimiza
tion technique with inner iterations of any optimization algorithm. The re
sulting NAM algorithmic framework is proved to converge to points that sat
isfy the first-order condition for optimality. This convergence guarantee
is the state-of-the-art in recent years\, due to non-convexity of such pro
blems. Central to the proof of the NAM convergence guarantee\, is a new ge
neralization of classical proof techniques in the non-convex setting\, tha
t allows for errors in accommodating the conditions for convergence. To th
e best of our knowledge\, this convergence guarantee is the first of this
kind for nested non-descent methods in the non-convex and non-smooth setti
ng.\n\nIn this thesis\, the proposed NAM framework is also applied to chal
lenging non-convex and non-smooth optimization problems\, including image
deblurring\, unsynchronized source localization\, and wireless sensor loca
lization. In addition\, we also provide a new procedure that prevents opti
mization algorithms that tackle the multi-variate localization problem fro
m being trapped in convergence points of inferior quality\, thus allowing
convergence to points with a lower and better function value.\n\n \;\n
\nZoom Link
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:378@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230723T130000
DTEND;TZID=Asia/Jerusalem:20230723T133000
DTSTAMP:20230625T112136Z
URL:https://dds.technion.ac.il/iemevents/learning-with-economic-and-social
-constraints-in-recommendation-systems/
SUMMARY:Learning with Economic and Social Constraints in Recommendation Sys
tems [ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Rotem Torkan \n Advisors: Omer Ben-Porat \n Where:
Bloomfield 424 From:\nTechnion\nAbstract:\n\nRecommendation systems are
dynamic economic systems that balance the needs of multiple stakeholders.
A recent line of work studies incentives from the content providers' point
of view. Content providers\, e.g.\, vloggers and bloggers\, contribute fr
esh content and rely on user engagement to create revenue and finance thei
r operations. In our work\, we propose a contextual multi-armed bandit set
ting to model the dependency of content providers on exposure. In our mode
l\, the system receives a user context in every round and has to select on
e of the arms. Every arm is a content provider who must receive a minimum
number of pulls every fixed time period (e.g.\, a month) to remain viable
in later rounds\; otherwise\, the arm departs and is no longer available.
The system aims to maximize the users' (content consumers) welfare. To tha
t end\, it should learn which arms are vital and ensure they remain viable
by subsidizing arm pulls if needed. We develop algorithms with sub-linear
regret\, as well as a lower bound that demonstrates that our algorithms a
re optimal up to logarithmic factors.\n\nAdditionally\, we study a follow-
up model where arms do not depart but rather produce content on demand. Na
mely\, each arm has a price\, and the learner\, equipped with a limited bu
dget\, can pick the set of arms that produce content in a given round. We
present preliminary results of both inefficient algorithms with optimal re
gret guarantees and efficient algorithms with sublinear regret guarantees
(although not optimal).
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:382@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230801T093000
DTEND;TZID=Asia/Jerusalem:20230801T100000
DTSTAMP:20230726T061822Z
URL:https://dds.technion.ac.il/iemevents/towards-a-provable-approximation-
to-the-perspective-n-point-problem/
SUMMARY:Towards a Provable Approximation to The Perspective-n-point Problem
[ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc candidate Roy Matza\n Advisors: Prof. Dan Feldman\, Pr
of. Yair Goldberg\n Where: Room 527\, Bloomfield building From:\nTechnion\
nAbstract:\n\nIn the Perspective-n-Point (PnP) problem\, we are given a 3D
model ("cloud points") and its 2D-image that was captured from a camera.
The goal is to estimate the pose (translation and rotation) of the camera.
More formally and generally\, given a set $P\\subseteq\\R^d$ of $n$ point
s (the cloud) and a set of $n$ lines\, each intersects the origin (the cam
era pinhole)\, we wish to compute $OPT$ which is the minimal sum of square
d of point-line distances.\n\nExisting solutions with provable error bound
s are unstable and highly exponential in the $d$\, which makes them comple
tely impractical\, even for the common case $d=3$. Hence\, over the years
numerous heuristics were suggested with arbitrarily large errors\, that ma
y fail to return any solution at all.\n\nWe suggest the first polynomial-t
ime algorithm in the input's size that yields a constant factor approximat
ion to $OPT$. For constant $d$ (as in PnP)\, the running time is linear in
the input's size. This is by forging links to polynomial system solving f
rom Geometric Algebra\, Clifford Torus\, core-sets from Computational Geom
etry\, and Fourier Series from Harmonic Analysis.\n\nExtensive experimenta
l results show that\, unlike competitive heuristics\, the approximation er
ror of the algorithm is indeed bounded.\n\nWe expect that our technique wi
ll be generalized to many other computer vision problems. In particular\,
we prove that the algorithm can easily be generalized to minimizing sum of
(non-squared) distances between the line and the points\, and many other
estimators that are more robust to outliers.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:383@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230817T110000
DTEND;TZID=Asia/Jerusalem:20230817T120000
DTSTAMP:20230725T071355Z
URL:https://dds.technion.ac.il/iemevents/enhanced-meta-label-correction-fo
r-coping-with-label-corruption/
SUMMARY:Enhanced Meta Label Correction for Coping with Label Corruption [ \
n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc Mitchell Keren Taraday\n Advisors: Dr. Chaim Baskin\n
Where: Taub 401 From:\nTechnion\nAbstract:\n\nTraditional methods for lear
ning with the presence of noisy labels have successfully handled datasets
with artificially injected noise but still fall short of adequately handli
ng real-world noise. With the increasing use of meta-learning in the diver
se fields of machine learning\, researchers leveraged auxiliary small clea
n datasets to meta-correct the training labels. Nonetheless\, existing met
a-label correction approaches are not fully exploiting their potential.\n\
nIn this study\, we propose an Enhanced Meta Label Correction approach abb
reviated as EMLC for the learning with noisy labels (LNL) problem. We re-e
xamine the meta-learning process and introduce faster and more accurate me
ta-gradient derivations. We propose a novel teacher architecture tailored
explicitly to the LNL problem\, equipped with novel training objectives. E
MLC outperforms prior approaches and achieves state-of-the-art results in
all standard benchmarks. Notably\, EMLC enhances the previous art on the n
oisy real-world dataset Clothing1M by 1.52% while requiring x0.5 the time
per epoch and with much faster convergence of the meta-objective when comp
ared to the baseline approach.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:384@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230823T130000
DTEND;TZID=Asia/Jerusalem:20230823T133000
DTSTAMP:20230809T062138Z
URL:https://dds.technion.ac.il/iemevents/future-aware-policies-minimizes-p
rospective-antibiotic-resistance-in-urinary-tract-infections/
SUMMARY:Future-Aware Policies Minimizes Prospective Antibiotic Resistance I
n Urinary Tract Infections [ \n Graduate Student Seminar\n Seminar
s\n \n ]
DESCRIPTION:By: M.Sc Ido Terner\n Advisors: Roy Kishony\n Where: Faculty of
Biology\, Auditorium From:\nTechnion\nAbstract: The current treatment par
adigm for bacterial infections focuses on choosing antibiotics based on th
e susceptibility of the existing infecting pathogen. Indeed\, in recent ye
ars extensive work was focused on developing experimental and computationa
l methodologies for rapidly measuring or estimating the resistance profile
of an infection as a basis for matching antibiotics to which the pathogen
is susceptible. However\, antibiotics are a double-edged sword: while the
y help clear the current infection\, they potentially select future-resist
ant pathogens that are harder to treat. Hence\, tailoring antibiotics for
a given infection at the single patient level should not just focus on cle
arance of the focal infection (“greedy”)\, but also on reducing the ri
sk of future resistant infections (“future-aware”). To address this ch
allenge\, we propose a multi-step treatment approach to learning personali
zed treatment policies using the framework of reinforcement learning (RL).
We tackle this problem in Urinary tract infections (UTIs)\, one of the mo
st common bacterial infections with high resistance prevalence. Harnessing
a unique nationwide cohort of over 100\,000 patients who had one or more
UTIs\, we develop a think-ahead model for optimizing current treatment bas
ed on its effect on both current and future clinical outcomes. Comparing o
ur optimal future-looking prescription model with a “greedy” model tha
t optimizes treatment of just the immediate focal infection\, we found tha
t a look-ahead strategy can much improve long-term treatment success. Our
work provides a proof-of-concept for using RL to learn treatment policies
that not only focus on treating the existing infection but also reduce the
burden of future ones.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:385@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230911T123000
DTEND;TZID=Asia/Jerusalem:20230911T133000
DTSTAMP:20230907T110048Z
URL:https://dds.technion.ac.il/iemevents/the-interplay-between-public-medi
a-news-sentiment-and-implied-volatility-of-returns/
SUMMARY:The Interplay between Public Media News Sentiment and Implied Volat
ility of Returns [ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Farid Srouji\n Advisors: Doron Kliger\n Where: Unive
rsity of Haifa\, Rabin Building\, 7 floor\, room 7052 From:\nTechnion/Un
iversity of Haifa\nAbstract:\nThe topic of Public Media News Sentiment and
Implied Volatility of Returns refers to the relationship between news sen
timent and stock market volatility. The stock market is influenced by a va
riety of factors\, including economic indicators\, political events\, and
news sentiment. Public media news sentiment refers to the overall tone and
sentiment of news stories reported by the mainstream media. The purpose o
f this research is to examine the relationship between public media news s
entiment and the implied volatility of stock market returns. Implied volat
ility refers to the expected level of volatility of a stock or index over
a given period.\nThe implied volatility index\, VIX\, is a widely recogniz
ed indicator utilized by investors and traders to evaluate the perceived r
isk or uncertainty associated with the financial markets. The study of thi
s topic is important because it helps investors understand how news sentim
ent can affect market volatility and risk. By understanding this relations
hip\, investors can make more informed investment decisions and manage the
ir risk exposure more effectively. Several studies have examined the relat
ionship between public media news sentiment and implied volatility. These
studies have found that news sentiment can have a significant impact on ma
rket volatility. For example\, positive news sentiment can lead to lower l
evels of volatility\, while negative news sentiment can lead to higher lev
els of volatility.\nOne way that investors can use this information is by
monitoring news sentiment and adjusting their investment strategies accord
ingly. For example\, if the news sentiment is negative\, investors may cho
ose to reduce their exposure to high-risk assets and increase their exposu
re to low-risk assets. In conclusion\, the topic of Public Media News Sent
iment and Implied Volatility of Returns is an important area of study for
investors and financial professionals. By understanding the relationship b
etween news sentiment and market volatility\, investors can make more info
rmed investment decisions and manage their risk exposure more effectively.
\nIn this thesis\, I leverage the power of natural language processing and
Python's textual analysis tools and generate sentiment time series from b
oth The New York Times and the S&\;P500 companies' press releases\, cov
ering the period from 1 June 2020 to 1 June 2023. Each source provides aro
und fifty thousand articles. The aim of this thesis is to uncover\, throug
h empirical work\, the potential of news sentiment of providing valuable i
nsights into the relationship with the implied volatility of returns.\nThe
empirical analysis consists of event-study\, in section three\, I use the
news volume to select the events days by choosing a threshold above or be
low the mean of the news volume abnormal returns. In section four\, I appl
ied and trained machine learning forecasting models based on the sentiment
returns as input fit to predict the direction of the VIX index returns. T
he forecasting models with the sentiment timeseries as input\, exhibit an
advantage in the prediction of the VIX direction compared to a random samp
le\, moreover\, I evaluated and ranked the models.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:389@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230914T103000
DTEND;TZID=Asia/Jerusalem:20230914T113000
DTSTAMP:20230904T050601Z
URL:https://dds.technion.ac.il/iemevents/quantifying-the-complex-interacti
ons-between-spaces-people-and-activities-in-built-environments-a-network-s
cience-approach/
SUMMARY:Quantifying the complex interactions between spaces\, people and ac
tivities in built Environments: a Network Science Approach [ \n Gradua
te Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Shaked Fried \n Advisors: Yoed Kenett\, Davide Sch
aumann \n Where: Bloomfield 424 From:\nTechnion\nAbstract:\n\nThe built
environment and especially mission-critical facilities such as healthcare
centers are complex entities in constant change. Proficient use of often-l
imited resources in operation-oriented environments may be gained by gener
ating a more holistic understanding of the built environment as an ecosyst
em of reciprocities between space\, people\, and activities.To tackle the
challenge of achieving seamless integration representation\, our proposal
involves the application of graph theory to explore the dynamics of these
systems.\n\nTogether with Rambam’s Institute of Pain Medicine as our cas
e study\, we’ll present a graph theory analysis\, commonly known as netw
ork science\, to decipher how hospital spaces\, resources\, and patient jo
urneys connect and can converge in a system-level representation. This uni
que framework has successfully facilitated predictive models\, unraveling
dynamics within the built environment system. Our findings provide insight
s into how shared resource patterns interact with service capacity\, space
utilization\, and more. This interdisciplinary approach sheds light on th
e intricate dynamics of built environments and paves the way on the journe
y toward intelligent environments.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:376@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20230919T133000
DTEND;TZID=Asia/Jerusalem:20230919T140000
DTSTAMP:20230816T070808Z
URL:https://dds.technion.ac.il/iemevents/detecting-higher-order-phenomena-
in-spatial-transcriptomics-through-topological-graph-neural-networks/
SUMMARY:Detecting Higher Order Phenomena in Spatial Transcriptomics Through
Topological Graph Neural Networks [ \n Graduate Student Seminar\n
Seminars\n \n ]
DESCRIPTION:By: M.Sc Oren Ploznik \n Advisors: Dr. Dvir Aran\n Where: Facul
ty of Biology\, Auditorium From:\nTechnion\nAbstract:\n\nSpatially Resolve
d Transcriptomics (SRT) technologies offer a new lens to study gene expres
sion profiles and spatial positioning of cells within tissues. However\, t
he integration of spatial information with gene expression data and deciph
ering complex phenomena\, such as cell-cell interactions\, remain challeng
ing. Existing strategies\, such as Graph Neural Networks\, create cell rep
resentations by blending gene expression and spatial context\, but they ca
n inadvertently introduce noise and inconsistencies\, thereby undermining
the robustness of the conclusions. In response to these challenges\, we in
troduce the Graph-based Higher Order Spatial Transcriptomics (GHOST) analy
sis toolkit. GHOST enables the efficient representation of individual cell
s and their spatial context within higher-order structures in an unsupervi
sed manner. It further incorporates functional reasoning tools for downstr
eam analysis. Using CW Networks and an enhanced Deep Graph Infomax framewo
rk\, GHOST generates representations not just for individual cells\, but a
lso for higher-order structures like cell communities that meet user-defin
ed criteria. This dual representation empowers researchers to efficiently
interpret the spatial context of these structures and to integrate it in e
ase with other methods. GHOST streamlines various analytical tasks includi
ng segmentation\, higher-order motif detection\, and the identification of
statistically significant interaction regions across tissues. Its effecti
veness is validated across multiple tissue types\, such as mouse hippocamp
us\, human brain metastases\, and human prostate cancer\, and through the
successful detection of higher-order phenomena in synthetic SRT samples. I
n essence\, GHOST is a powerful tool that enables efficient representation
and interpretation of both lower and higher-order biological phenomena in
spatial context\, thus enriching our understanding of intricate biologica
l processes in SRT data.
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:394@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20231023T113000
DTEND;TZID=Asia/Jerusalem:20231023T120000
DTSTAMP:20231010T122613Z
URL:https://dds.technion.ac.il/iemevents/b-learner-quasi-oracle-bounds-on-
heterogeneous-causal-effects-and-policy-learning-from-observational-data-w
ith-the-option-of-deferring-to-an-expert-under-hidden-confounding/
SUMMARY:B-Learner: Quasi-Oracle Bounds on Heterogeneous Causal Effects\, an
d Policy Learning from observational data with the option of deferring to
an Expert Under Hidden Confounding [ \n Graduate Student Seminar\n
Seminars\n \n ]
DESCRIPTION:By: M.Sc. Marah Ghoummaid \n Advisors: Uri Shalit\n Where: ZOO
M From:\nTechnion\nAbstract:\n\nEstimating heterogeneous treatment effects
from observational data is a crucial task across many fields\, helping po
licy and decision-makers take better actions. There has been recent progre
ss on robust and efficient methods for estimating the conditional average
treatment effect (CATE) function\, but these methods often do not take int
o account the risk of hidden confounding\, which could arbitrarily and unk
nowingly bias any causal estimate based on observational data. We propose
a meta-learner called the B-Learner\, which can efficiently learn sharp bo
unds on the CATE function under limits on the level of hidden confounding.
We derive the B-Learner by adapting recent results for sharp and valid bo
unds of the average treatment effect (Dorn et al.\, 2021) into the framewo
rk given by Kallus &\; Oprescu (2023) for robust and model-agnostic lea
rning of conditional distributional treatment effects. The B-Learner can u
se any function estimator such as random forests and deep neural networks\
, and we prove its estimates are valid\, sharp\, efficient\, and have a qu
asi-oracle property with respect to the constituent estimators under more
general conditions than existing methods. Semi-synthetic experimental comp
arisons validate the theoretical findings\, and we use real-world data to
demonstrate how the method might be used in practice.\n\nIn addition\, we
then make use of our B-Learner estimator for a crucial task: Policy learni
ng from observational data where the policy is being learned in conjunctio
n with an expert\, i.e.\, the task is to learn treatment assignment where
the model can either assign a treatment\, or defer the decision to an expe
rt. This task is especially challenging when dealing with observational da
ta\, as we in most cases face the risk of having hidden confounders\, whic
h if not being accounted for\, can lead to harmful policies. Our work buil
ds on previous work off Mozannar and Sontag (2020) on consistent surrogate
loss for learning with the option of deferral to an expert\, where they s
olve a cost-sensitive supervised classification problem. Since we are solv
ing a causal problem\, where labels don’t exist\, we use the B-Learner e
stimator to build a consistent loss for our model to learn a policy that c
an work in conjunction with an expert. Our model can take advantage of t
he strengths points of both the model and the expert to obtain a better po
licy\, which is\, in addition\, robust to hidden confounders.\n\nZoom Link
\n\nhttps://technion.zoom.us/j/97950728015
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VEVENT
UID:395@dds.technion.ac.il
DTSTART;TZID=Asia/Jerusalem:20231122T103000
DTEND;TZID=Asia/Jerusalem:20231122T113000
DTSTAMP:20231116T131810Z
URL:https://dds.technion.ac.il/iemevents/secure-and-adaptive-parallel-opti
mization-a-history-independent-approach/
SUMMARY:Secure and Adaptive Parallel Optimization: A History Independent Ap
proach [ \n Graduate Student Seminar\n Seminars\n \n ]
DESCRIPTION:By: M.Sc. Naseem Yehya\n Advisors: Dr. Kfir Levi \n Where: ZO
OM From:\nTechnion\nAs machine learning continues to play an integral role
in numerous applications\, the robustness and security of distributed mac
hine learning models\, particularly in the presence of Byzantine failures\
, have emerged as critical research areas. This thesis presents a novel ap
proach to Byzantine fault tolerance\, specifically targeting the dependenc
e on gradient history. We introduce an adaptive Byzantine machine learning
model that leverages Multi-Level Monte Carlo (MLMC) methods\, implemented
on the RenNet9 and LeNet deep neural network\, to classify the CIFAR-10 a
nd MNIST dataset. \n Our research is distinguished by several key finding
s. Unlike conventional approaches\, our model operates independently of th
e knowledge of the percentage of Byzantine workers\, making it adaptable t
o dynamic scenarios where Byzantine nodes may change over time—an aspect
often overlooked in existing literature. We emphasize practicality by dem
onstrating both theoretical convergence proofs and a robust implementation
\, with a specific focus on real-world applicability.\n The keywords "MLM
C\," "SGD\," "Byzantine\," "Convergence\," "deep learning simulation\," an
d "Adaptive to Byzantine" encapsulate the core elements of our research. I
n our approach\, we first establish the theoretical foundations\, presenti
ng rigorous proofs and insights into the model's adaptability and converge
nce properties. Subsequently\, we transition to implementation\, highlight
ing practical considerations and providing evidence of our model's effecti
veness.\n This research contributes to the advancement of Byzantine fault
-tolerant machine learning by introducing a history-independent\, adaptive
approach that addresses dynamic Byzantine worker scenarios. Our model not
only extends the theoretical understanding of Byzantine fault tolerance b
ut also showcases its viability in real-world deep learning applications.
By bridging the gap between theory and practice\, our work offers valuable
insights into the development of robust and adaptive machine learning mod
els for distributed environments.\nZOOM Link\n\nhttps://technion.zoom.us/j
/94587746996
CATEGORIES:Graduate Student Seminar,Seminars
END:VEVENT
BEGIN:VTIMEZONE
TZID:Asia/Jerusalem
X-LIC-LOCATION:Asia/Jerusalem
BEGIN:STANDARD
DTSTART:20191027T010000
TZOFFSETFROM:+0300
TZOFFSETTO:+0200
TZNAME:IST
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20200327T030000
TZOFFSETFROM:+0200
TZOFFSETTO:+0300
TZNAME:IDT
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:20201025T010000
TZOFFSETFROM:+0300
TZOFFSETTO:+0200
TZNAME:IST
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20210326T030000
TZOFFSETFROM:+0200
TZOFFSETTO:+0300
TZNAME:IDT
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:20211031T010000
TZOFFSETFROM:+0300
TZOFFSETTO:+0200
TZNAME:IST
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20220325T030000
TZOFFSETFROM:+0200
TZOFFSETTO:+0300
TZNAME:IDT
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:20221030T010000
TZOFFSETFROM:+0300
TZOFFSETTO:+0200
TZNAME:IST
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20230324T030000
TZOFFSETFROM:+0200
TZOFFSETTO:+0300
TZNAME:IDT
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:20231029T010000
TZOFFSETFROM:+0300
TZOFFSETTO:+0200
TZNAME:IST
END:STANDARD
END:VTIMEZONE
END:VCALENDAR