End-to-end Learning of Compressible Features

TL;DR

This paper introduces a learned, end-to-end method for producing features from CNNs that are highly compressible and improve task accuracy, outperforming traditional compression techniques.

Contribution

The authors propose a novel joint optimization approach for feature compressibility and task performance, adaptable to any target objective.

Findings

Features are an order of magnitude more compressible.

Improved accuracy due to regularization effect.

Effective across multiple benchmarks.

Abstract

Pre-trained convolutional neural networks (CNNs) are powerful off-the-shelf feature generators and have been shown to perform very well on a variety of tasks. Unfortunately, the generated features are high dimensional and expensive to store: potentially hundreds of thousands of floats per example when processing videos. Traditional entropy based lossless compression methods are of little help as they do not yield desired level of compression, while general purpose lossy compression methods based on energy compaction (e.g. PCA followed by quantization and entropy coding) are sub-optimal, as they are not tuned to task specific objective. We propose a learned method that jointly optimizes for compressibility along with the task objective for learning the features. The plug-in nature of our method makes it straight-forward to integrate with any target objective and trade-off against compressibility. We present results on multiple benchmarks and demonstrate that our method produces features that are an order of magnitude more compressible, while having a regularization effect that leads to a consistent improvement in accuracy.