EAST: Encoding-Aware Sparse Training for Deep Memory Compression of ConvNets

TL;DR

EAST is a novel training method that enhances deep memory compression of ConvNets by adaptive pruning, achieving better accuracy at lower sparsity levels on resource-constrained devices.

Contribution

The paper introduces EAST, a memory-aware sparse training approach that improves compression efficiency and accuracy of quantized ConvNets under strict memory constraints.

Findings

EAST achieves higher accuracy with lower sparsity compared to existing methods.

EAST successfully compresses ResNet-9 on a microcontroller with minimal accuracy loss.

Adaptive group pruning maximizes compression rate of weight encoding schemes.

Abstract

The implementation of Deep Convolutional Neural Networks (ConvNets) on tiny end-nodes with limited non-volatile memory space calls for smart compression strategies capable of shrinking the footprint yet preserving predictive accuracy. There exist a number of strategies for this purpose, from those that play with the topology of the model or the arithmetic precision, e.g. pruning and quantization, to those that operate a model agnostic compression, e.g. weight encoding. The tighter the memory constraint, the higher the probability that these techniques alone cannot meet the requirement, hence more awareness and cooperation across different optimizations become mandatory. This work addresses the issue by introducing EAST, Encoding-Aware Sparse Training, a novel memory-constrained training procedure that leads quantized ConvNets towards deep memory compression. EAST implements an adaptive group pruning designed to maximize the compression rate of the weight encoding scheme (the LZ4 algorithm in this work). If compared to existing methods, EAST meets the memory constraint with lower sparsity, hence ensuring higher accuracy. Results conducted on a state-of-the-art ConvNet (ResNet-9) deployed on a low-power microcontroller (ARM Cortex-M4) validate the proposal.