Lossless compression with state space models using bits back coding

James Townsend, Iain Murray

Keywords: variational inference, bits-back coding, compression, lossless compression, latent variables, approximate inference, state space models, hidden Markov models

TL;DR: We show how to do bits-back coding with a general class of state space models, demonstrate the idea on a hidden Markov model

Abstract: We generalize the 'bits back with ANS' method to time-series models with a latent Markov structure. This family of models includes hidden Markov models (HMMs), linear Gaussian state space models (LGSSMs) and many more. We provide experimental evidence that our method is effective for small scale models, and discuss its applicability to larger scale settings such as video compression.

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/arxiv:2103.10150/code)