Lossy Compression for Lossless Prediction
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 1000x on ImageNet) compared to JPEG on 8 datasets, without decreasing downstream classification performance.
Cite
Text
Dubois et al. "Lossy Compression for Lossless Prediction." Neural Information Processing Systems, 2021.Markdown
[Dubois et al. "Lossy Compression for Lossless Prediction." Neural Information Processing Systems, 2021.](https://mlanthology.org/neurips/2021/dubois2021neurips-lossy/)BibTeX
@inproceedings{dubois2021neurips-lossy,
title = {{Lossy Compression for Lossless Prediction}},
author = {Dubois, Yann and Bloem-Reddy, Benjamin and Ullrich, Karen and Maddison, Chris J},
booktitle = {Neural Information Processing Systems},
year = {2021},
url = {https://mlanthology.org/neurips/2021/dubois2021neurips-lossy/}
}