Over-the-Air Machine Learning at the Wireless Edge

TLDR

The paper investigates over‑the‑air distributed stochastic gradient descent for power‑constrained edge devices, using analog transmission of low‑dimensional gradients and error accumulation across iterations. The authors model a bandwidth‑limited fading multiple‑access channel, design an opportunistic worker‑scheduling scheme to align received gradients, and employ analog transmission to aggregate gradients at a remote parameter server. Numerical experiments demonstrate that the proposed DSGD algorithm converges faster and achieves higher accuracy than state‑of‑the‑art methods.

Abstract

We study distributed machine learning at the wireless edge, where limited power devices (workers) with local datasets implement distributed stochastic gradient descent (DSGD) over-the-air with the help of a remote parameter server (PS). We consider a bandwidth-limited fading multiple access channel (MAC) from the workers to the PS for communicating the local gradient estimates. Motivated by the additive nature of the wireless MAC, we study analog transmission of low-dimensional gradient estimates while accumulating error from previous iterations. We also design an opportunistic worker scheduling scheme to align the received gradient vectors at the PS in an efficient manner. Numerical results show that the proposed DSGD algorithm converges much faster than the state-of-the-art, while also providing a significantly higher accuracy.