Weightless: Lossy Weight Encoding For Deep Neural Network Compression

TL;DR

This paper introduces Weightless, a lossy weight encoding scheme using Bloomier filters that significantly compresses neural network weights while maintaining accuracy, surpassing current methods.

Contribution

The novel use of Bloomier filters for lossy weight encoding in neural networks enables unprecedented compression ratios with minimal accuracy loss.

Findings

Achieves up to 496x compression of DNN weights.

Maintains model accuracy after compression.

Outperforms state-of-the-art compression techniques by 1.51x.

Abstract

The large memory requirements of deep neural networks limit their deployment and adoption on many devices. Model compression methods effectively reduce the memory requirements of these models, usually through applying transformations such as weight pruning or quantization. In this paper, we present a novel scheme for lossy weight encoding which complements conventional compression techniques. The encoding is based on the Bloomier filter, a probabilistic data structure that can save space at the cost of introducing random errors. Leveraging the ability of neural networks to tolerate these imperfections and by re-training around the errors, the proposed technique, Weightless, can compress DNN weights by up to 496x with the same model accuracy. This results in up to a 1.51x improvement over the state-of-the-art.