Weightless: Lossy Weight Encoding For Deep Neural Network Compression
Published in
International Conference on Machine Learning(2017)External Links:
TL
TL;DR
A novel scheme for lossy weight encoding co-designed with weight simplification techniques that can compress weights by up to 496x without loss of model accuracy, resulting in up to a 1.51x improvement over the state-of-the-art.
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 co-designed with weight simplification 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, named Weightless, can compress weights by up to 496x without loss of model accuracy. This results in up to a 1.51x improvement over the state-of-the-art.
References30 items
1
Exploring Sparsity in Recurrent Neural Networks
2
Soft Weight-Sharing for Neural Network Compression
3
Towards the Limit of Network Quantization
4
Training Compressed Fully-Connected Networks with a Density-Diversity Penalty
5