Lossy Compression for Lossless Prediction
This work addresses the inefficiency of traditional compression for algorithm-driven data analysis, offering a novel approach that could reduce storage and transmission costs in machine learning applications.
The paper tackles the problem of designing data compressors that preserve only the information needed for downstream predictive tasks, rather than perceptual fidelity, and demonstrates a neural compressor achieving over 1000x rate savings on ImageNet compared to JPEG without harming classification performance across 8 datasets.
Most data is automatically collected and only ever "seen" by algorithms. Yet, data compressors preserve perceptual fidelity rather than just the information needed by algorithms performing downstream tasks. In this paper, we characterize the bit-rate required to ensure high performance on all predictive tasks that are invariant under a set of transformations, such as data augmentations. Based on our theory, we design unsupervised objectives for training neural compressors. Using these objectives, we train a generic image compressor that achieves substantial rate savings (more than $1000\times$ on ImageNet) compared to JPEG on 8 datasets, without decreasing downstream classification performance.