RCT: Resource Constrained Training for Edge AI

TL;DR

This paper introduces Resource Constrained Training (RCT), a method for training quantised neural networks on edge devices that reduces memory and energy consumption by maintaining a quantised model and dynamically adjusting bitwidths during training.

Contribution

RCT is a novel training approach that minimizes resource usage on edge devices by keeping models quantised and adapting precision per layer, enabling efficient training under resource constraints.

Findings

RCT reduces energy consumption by over 86% for GEMM operations.

RCT cuts memory usage by more than 46% for model parameters.

Compared to QAT, RCT halves the energy required for model parameter movement.

Abstract

Neural networks training on edge terminals is essential for edge AI computing, which needs to be adaptive to evolving environment. Quantised models can efficiently run on edge devices, but existing training methods for these compact models are designed to run on powerful servers with abundant memory and energy budget. For example, quantisation-aware training (QAT) method involves two copies of model parameters, which is usually beyond the capacity of on-chip memory in edge devices. Data movement between off-chip and on-chip memory is energy demanding as well. The resource requirements are trivial for powerful servers, but critical for edge devices. To mitigate these issues, We propose Resource Constrained Training (RCT). RCT only keeps a quantised model throughout the training, so that the memory requirements for model parameters in training is reduced. It adjusts per-layer bitwidth dynamically in order to save energy when a model can learn effectively with lower precision. We carry out experiments with representative models and tasks in image application and natural language processing. Experiments show that RCT saves more than 86\% energy for General Matrix Multiply (GEMM) and saves more than 46\% memory for model parameters, with limited accuracy loss. Comparing with QAT-based method, RCT saves about half of energy on moving model parameters.