On the advantages of stochastic encoders

Lucas Theis, Eirikur Agustsson

Keywords: compression, autoencoders, universal quantization, perception-distortion trade-off, realism

TL;DR: An example showing that stochastic encoders can outperform deterministic encoders when optimizing realism in addition to distortion

Abstract: Stochastic encoders have been used in rate-distortion theory and neural compression because they can be easier to handle. However, in performance comparisons with deterministic encoders they often do worse, suggesting that noise in the encoding process may generally be a bad idea. It is poorly understood if and when stochastic encoders can do better than deterministic encoders. In this paper we provide one illustrative example which shows that stochastic encoders can significantly outperform the best deterministic encoders. Our toy example suggests that stochastic encoders may be particularly useful in the regime of "perfect perceptual quality".