Download as iCal file

First-Order Methods for Solving the Sensor Network Localization Problem

By Eyal Gur
Location Bloomfield 527
Advisor(s): Dr. Shoham Sabach, Dr. Shimrit Shtern
MSc
Academic Program: OR
 
Monday 08 July 2019, 14:30 - 15:30
A wireless sensor network consists of a number of sensors deployed in a given area, for purposes such as environment monitoring, battlefields surveillance and more. Each sensor is a low-powered radio transceiver which monitors its immediate surroundings (e.g., for temperature, sound, etc.), and the data is collected by a processor. Therefore, the location of each sensor has a significant role. Since the amount of sensors in a network can be large (even thousands of sensors), it is not cost effective to equip each of them with a GPS device, nor to deploy the sensors in a known logged location. This is the main motivation to introduce sensor location estimation techniques.
We formulate the the sensor localization problem as a non-smooth non-convex optimization model. We develop several first-order iterative algorithms, which all can be applied in a parallel or a centralized configuration (whether a central processor is required or not) and with a synchronous/asynchronous updating rules.
The novelty in our techniques, is that to the best of our knowledge, we first establish a globally convergent, computationally efficient and a fully parallelizable algorithms, that are based on a non-relaxed model formulation, which can be applied to a large-scale networks.