Efficient Neural Compression with Inference-time Decoding

TL;DR

This paper presents a novel neural compression method combining mixed precision, zero-point quantization, and entropy coding, enabling efficient ResNet model compression with minimal accuracy loss and inference-time decoding capabilities.

Contribution

It introduces a new approach that surpasses the 1-bit quantization frontier, achieving high compression rates with less than 1% accuracy loss on ImageNet.

Findings

Achieves compression beyond the 1-bit frontier.

Maintains less than 1% accuracy loss on ImageNet.

Enables inference-compatible decoding with reduced latency.

Abstract

This paper explores the combination of neural network quantization and entropy coding for memory footprint minimization. Edge deployment of quantized models is hampered by the harsh Pareto frontier of the accuracy-to-bitwidth tradeoff, causing dramatic accuracy loss below a certain bitwidth. This accuracy loss can be alleviated thanks to mixed precision quantization, allowing for more flexible bitwidth allocation. However, standard mixed precision benefits remain limited due to the 1-bit frontier, that forces each parameter to be encoded on at least 1 bit of data. This paper introduces an approach that combines mixed precision, zero-point quantization and entropy coding to push the compression boundary of Resnets beyond the 1-bit frontier with an accuracy drop below 1% on the ImageNet benchmark. From an implementation standpoint, a compact decoder architecture features reduced latency, thus allowing for inference-compatible decoding.