Robust Sequential Detection via Robust Estimation in Sensor Networks
Masterarbeit, Proseminar, Studienarbeit, Bachelorarbeit
The reliable detection of events in a network of intelligent (and possibly heterogeneous) sensors is a ubiquitous, yet challenging task with applications in home automation, earthquake or forest fire detection, air and ground traffic control, cognitive radio, video surveillance, etc.Sequential detection is an important field of research that seeks to evaluate data sequentially as it comes in, instead of collecting a predefined number of samples before making a decision. That way, events can be detected much faster but care must be taken to guarantee reliable decisions with minimal false alarms and missed detections.
In order to develop algorithms that solve the detection problem optimally and in real-time, assumptions are made based on the application at hand. If these assumptions are violated, for instance, due to harsh environmental conditions or outliers in the data, these algorithms can break down. For this reason, it is essential to come up with robust algorithms that are reliable even in the face of deviating assumptions.
Students participating in this project will learn about detection and estimation theory with special emphasis on sequential methods in distributed setups as well as robust statistics. Distributed detection algorithms usually involve the collaborative estimation of a common test statistic before performing the detection task. In this project, students should investigate how to robustify this preceding estimation step rather than the detector itself in order to obtain robust algorithms for distributed sequential detection problems.