CSEE Seminar Series
Dr David Burth Kurka presents a talk on fascinating developments in joint source-channel coding through Machine Learning techniques
Almost all wireless communication systems today are designed based on essentially the same digital approach, that separately optimises the compression and channel coding stages. Using machine learning techniques, we investigate whether end-to-end transmission can be learned from scratch, thus using joint source-channel coding (JSCC) rather than the separation approach.
The talk will review recent developments and introduce deep-JSCC - a set of autoencoder-based solutions for generating robust and compact codes directly from image pixels. From a totally data-driven approach, deep-JSCC shows comparable or even superior performance compared to state-of-the-art separation schemes and present a series of important features, such as graceful degradation, versatility to different channels and domains, variable transmission rate and capability to exploit channel output feedback.
David Burth Kurka is a postdoc researcher at Information Processing and Communications Lab, Imperial College London, working with machine learning for wireless communications.
Prior to that, David completed his PhD at the Electrical and Electronic Department at Imperial College, and obtained MSc and BEng degrees at University of Campinas, Brazil. His research interests include machine learning, information theory, multi-agent-systems, socio-inspired computing, distributed intelligence and complex systems.