Research seminar series - Distributed Optimization over Communication-constrained Networks: Enabling the Next Generation Machine Intelligence

#Distributed #optimization #decentralized #computational #strategy #wireless #communication

Research seminar series organized by Sensors Council, Vehicular Technology Society, and Communications Society Singapore Chapter

In today's networked world, enormous amounts of data are being collected by edge devices in a distributed fashion. It is often preferred to keep the data localized on the edge devices themselves for privacy concerns, fault-resilience, or the sheer communication expense of centrally aggregating the available data. As a consequence, there has been a rapid emergence of decentralized computational strategies in which global learning tasks are accomplished only through local processing. It this talk we shall review some topics in the decentralized optimization setting in which multiple nodes in a network collaborate to minimize the sum of their local loss functions.

We will consider both the centralized learning paradigm (a.k.a. federated learning) and the fully distributed paradigm. In both cases, we will introduce the relevant communication constraint for instance (i) a finite rate-constraint on the signal transmitted by any node, (ii) additive noise corrupting the signal received by any node, and (iii) straggling due to networking. We will review some of the techniques proposed in the literature to address such constraint, as well as some of my contributions on this topic.  The talk will conclude with some interesting research directions at the intersection of distributed optimization and wireless communication.