Distributed Signal Processing and Learning: Overview and Recent Developments

Distributed signal processing and learning (DSPL) over networks relies on in-network processing and cooperation among neighboring agents. Each agent performs some signal processing and learning task such as estimation, detection, compression, model fitting, classification, clustering and so forth.  The area emerged more than ten years ago but the interest of the community continues to grow fast.  This is due to a number of reasons, such as:  a) the new interesting theoretical/algorithmic problems posed, b) the close connection with parallel developments in other areas (e.g. Graph Signal Processing), and c)  the wide potential applicability to several types of networks.

The talk will start by introducing distributed signal processing and learning techniques and will explore some indicative application areas, such as the Internet of Things, Machine-to-Machine communications, Spectrum Sensing and Direction of Arrival Estimation. In the sequel, various approaches for DSPL will be reviewed, in terms of the algorithms employed (diffusion adaptive, consensus, gossip, game theoretic), in terms of the considered scenario (single task, multi-task / node specific) and in terms of the assumptions made. Finally, the talk will focus on three more specific aspects, that is, node specific distributed parameter estimation and coalition formation, distributed dictionary learning, and the use of game theory to treat distributed parameter estimation in the presence of ambiguous measurements.