NSF Funded MLWins Project Website

Visit PI Institutional Research Pages for more related works: Prof. Z. Ding

Collaborative Research: MLWiNS: Distributed Learning over Multi-Access Channels: From Bandlimited Coordinate Descent to Gradient Sketching
Project Overview

The past decade has witnessed an explosive growth of Internet of Things (IoT) services. An earlier report by Cisco predicted that the number of connected devices and subjects (including machines, humans, and things) can potentially grow to 500 billions by 2025. A general consensus is that much of IoT-created data will have to be stored and analyzed near, or at the network edge. This need has given rise to a new computing paradigm, namely edge computing, which features a new architecture by extending cloud computing to the network edge with low latency. Aiming to develop an integrated distributed learning and wireless networking framework, this project takes a principled approach to investigate two categories of edge learning algorithms tailored for wireless MAC channels, namely bandlimited coordinate descent and bandlimited gradient sketching. In these algorithms, sparsified or sketched versions of local updates are exchanged using multicarrier transmissions constrained by bandwidth; and each sender carries out power allocation across subcarriers based on gradient values and channel conditions through coordinated integration Furthermore, we devise edge learning algorithms using two intimately related methods: 1) first-order stochastic gradient descent methods and 2) zero-order stochastic optimization methods. In particular, zero-order methods are derivative-free so they are very promising for wireless edge learning.