Lossy Compression for Lossless Prediction

arXiv2106.10800MaRDI QIDQ6370794

Chris J. Maddison, Karen Ullrich, Yann Dubois, Benjamin Bloem-Reddy

Publication date: 20 June 2021

Abstract: 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 $1000imes$ on ImageNet) compared to JPEG on 8 datasets, without decreasing downstream classification performance.

Has companion code repository: https://github.com/YannDubs/lossyless

This page was built for publication: Lossy Compression for Lossless Prediction

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6370794)