`

Timezone: »

 
Poster
Exact Rate-Distortion in Autoencoders via Echo Noise
Rob Brekelmans · Daniel Moyer · Aram Galstyan · Greg Ver Steeg

Tue Dec 10 10:45 AM -- 12:45 PM (PST) @ East Exhibition Hall B + C #67
in Algorithms -- Representation Learning »

Compression is at the heart of effective representation learning. However, lossy compression is typically achieved through simple parametric models like Gaussian noise to preserve analytic tractability, and the limitations this imposes on learning are largely unexplored. Further, the Gaussian prior assumptions in models such as variational autoencoders (VAEs) provide only an upper bound on the compression rate in general. We introduce a new noise channel, Echo noise, that admits a simple, exact expression for mutual information for arbitrary input distributions. The noise is constructed in a data-driven fashion that does not require restrictive distributional assumptions. With its complex encoding mechanism and exact rate regularization, Echo leads to improved bounds on log-likelihood and dominates beta-VAEs across the achievable range of rate-distortion trade-offs. Further, we show that Echo noise can outperform flow-based methods without the need to train additional distributional transformations.

Keywords: Information Theory · Unsupervised Learning · Theory · Algorithms -> Representation Learning; Algorithms

Author Information

Rob Brekelmans (University of Southern California)
Daniel Moyer (MIT CSAIL)
Aram Galstyan (USC Information Sciences Institute)
Greg Ver Steeg (USC Information Sciences Institute)

More from the Same Authors