Alternating Information Bottleneck Method for Distributed Quantization, Prof. Volker Kuehn


11:00 - 12:30

EI 1 Petritsch Hörsaal

Loading Map....

Prof. Volker Kuehn from the University of Rostock, Germany will give a talk entitled

“Alternating Information Bottleneck Method for Distributed Quantization”

Distributed sensing systems with a large number of sensors will play an important role in modern communication systems. Industry 4.0 environments, the Internet of Things or environmental monitoring are just some examples for which massive machine-type communication has to be handled. Moreover, cloud and fog-based radio access networks can also be interpreted as distributed sensing systems. In all cases, limited communication resources require compression at the sensing nodes. As distributed sensors usually measure statistically depending signals, the sensor specific compression strategies shall be jointly optimized.

In this talk, a simple distributed sensing system is considered where the compression at each sensor node is performed by simple scalar quantization (clustering). While quantization is done individually at each sensor, the quantizers are jointly optimized such that individual rate constraints are fulfilled and the overall relevant mutual information is maximized. This optimization is performed using an adapted version of the information bottleneck principle. After briefly introducing the rate distortion theory and a generalization for noisy observations, the information bottleneck principle is introduced. For the distributed sensing system, an extended version is explained which is based on an alternating iterative algorithm. Numerical examples demonstrate the performance of the proposed approach.