Training on the Edge: The why and the how

TL;DR

This paper explores the advantages and challenges of performing machine learning training directly on edge devices, discussing scenarios where edge training is beneficial and strategies to optimize memory usage.

Contribution

It provides insights into when edge training is advantageous and introduces checkpointing strategies to manage memory constraints during training.

Findings

Edge training can be beneficial in privacy-sensitive scenarios.

Checkpointing strategies help reduce memory usage during edge training.

Certain applications benefit from decentralized training on edge devices.

Abstract

Edge computing is the natural progression from Cloud computing, where, instead of collecting all data and processing it centrally, like in a cloud computing environment, we distribute the computing power and try to do as much processing as possible, close to the source of the data. There are various reasons this model is being adopted quickly, including privacy, and reduced power and bandwidth requirements on the Edge nodes. While it is common to see inference being done on Edge nodes today, it is much less common to do training on the Edge. The reasons for this range from computational limitations, to it not being advantageous in reducing communications between the Edge nodes. In this paper, we explore some scenarios where it is advantageous to do training on the Edge, as well as the use of checkpointing strategies to save memory.