Variational Bayesian Quantization

We propose a novel algorithm for quantizing continuous latent representations in trained models. Our approach applies to deep probabilistic models, such as variational autoencoders (VAEs), and enables both data and model compression. Unlike current end-to-end neural compression methods that cater the model to a fixed quantization scheme, our algorithm separates model design and training from quantization. Consequently, our algorithm enables "plug-and-play" compression with variable rate-distortion trade-off, using a single trained model. Our algorithm can be seen as a novel extension of arithmetic coding to the continuous domain, and uses adaptive quantization accuracy based on estimates of posterior uncertainty. Our experimental results demonstrate the importance of taking into account posterior uncertainties, and show that image compression with the proposed algorithm outperforms JPEG over a wide range of bit rates using only a single standard VAE. Further experiments on Bayesian neural word embeddings demonstrate the versatility of the proposed method.

[1]  Valero Laparra,et al.  End-to-end Optimized Image Compression , 2016, ICLR.

[2]  Taco S. Cohen,et al.  Video Compression With Rate-Distortion Autoencoders , 2019, 2019 IEEE/CVF International Conference on Computer Vision (ICCV).

[3]  Jeffrey Dean,et al.  Distributed Representations of Words and Phrases and their Compositionality , 2013, NIPS.

[4]  Stephan Mandt,et al.  Improving Inference for Neural Image Compression , 2020, NeurIPS.

[5]  David M. Blei,et al.  Variational Inference: A Review for Statisticians , 2016, ArXiv.

[6]  David Minnen,et al.  Improved Lossy Image Compression with Priming and Spatially Adaptive Bit Rates for Recurrent Networks , 2017, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.

[7]  Lucas Theis,et al.  Lossy Image Compression with Compressive Autoencoders , 2017, ICLR.

[8]  Toby Berger Optimum quantizers and permutation codes , 1972, IEEE Trans. Inf. Theory.

[9]  D. A. Bell,et al.  Information Theory and Reliable Communication , 1969 .

[10]  Zhou Wang,et al.  Multiscale structural similarity for image quality assessment , 2003, The Thrity-Seventh Asilomar Conference on Signals, Systems & Computers, 2003.

[11]  Luca Benini,et al.  Soft-to-Hard Vector Quantization for End-to-End Learning Compressible Representations , 2017, NIPS.

[12]  Hedvig Kjellström,et al.  Advances in Variational Inference , 2017, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[13]  Christopher Burgess,et al.  beta-VAE: Learning Basic Visual Concepts with a Constrained Variational Framework , 2016, ICLR 2016.

[14]  Thomas M. Cover,et al.  Elements of Information Theory , 2005 .

[15]  Max Welling,et al.  Auto-Encoding Variational Bayes , 2013, ICLR.

[16]  Jeffrey Dean,et al.  Efficient Estimation of Word Representations in Vector Space , 2013, ICLR.

[17]  David J. C. MacKay,et al.  Information Theory, Inference, and Learning Algorithms , 2004, IEEE Transactions on Information Theory.

[18]  Alexander A. Alemi,et al.  Fixing a Broken ELBO , 2017, ICML.

[19]  David Minnen,et al.  Variable Rate Image Compression with Recurrent Neural Networks , 2015, ICLR.

[20]  David Barber,et al.  Practical Lossless Compression with Latent Variables using Bits Back Coding , 2019, ICLR.

[21]  Daan Wierstra,et al.  Towards Conceptual Compression , 2016, NIPS.

[22]  Stephan Mandt,et al.  Variable-Bitrate Neural Compression via Bayesian Arithmetic Coding , 2020, ICML 2020.

[23]  Antti Honkela,et al.  Variational learning and bits-back coding: an information-theoretic view to Bayesian learning , 2004, IEEE Transactions on Neural Networks.

[24]  Ian H. Witten,et al.  Arithmetic coding for data compression , 1987, CACM.

[25]  Daan Wierstra,et al.  Stochastic Backpropagation and Approximate Inference in Deep Generative Models , 2014, ICML.

[26]  David J. C. MacKay,et al.  A Practical Bayesian Framework for Backpropagation Networks , 1992, Neural Computation.

[27]  Luc Van Gool,et al.  Conditional Probability Models for Deep Image Compression , 2018, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.

[28]  Lubomir D. Bourdev,et al.  Real-Time Adaptive Image Compression , 2017, ICML.

[29]  S. P. Lloyd,et al.  Least squares quantization in PCM , 1982, IEEE Trans. Inf. Theory.

[30]  Michael I. Jordan,et al.  Latent Dirichlet Allocation , 2001, J. Mach. Learn. Res..

[31]  Julien Cornebise,et al.  Weight Uncertainty in Neural Networks , 2015, ArXiv.

[32]  D. Huffman A Method for the Construction of Minimum-Redundancy Codes , 1952 .

[33]  David Minnen,et al.  Variational image compression with a scale hyperprior , 2018, ICLR.

[34]  Stephan Mandt,et al.  Dynamic Word Embeddings , 2017, ICML.

[35]  Allen Gersho,et al.  Vector quantization and signal compression , 1991, The Kluwer international series in engineering and computer science.

[36]  Erez Lieberman Aiden,et al.  Quantitative Analysis of Culture Using Millions of Digitized Books , 2010, Science.

[37]  Stephan Mandt,et al.  Deep Generative Video Compression , 2018, NeurIPS.

[38]  Michael I. Jordan,et al.  An Introduction to Variational Methods for Graphical Models , 1999, Machine Learning.

[39]  Oren Barkan,et al.  Bayesian Neural Word Embedding , 2016, AAAI.

[40]  Sean Gerrish,et al.  Black Box Variational Inference , 2013, AISTATS.

[41]  Ruslan Salakhutdinov,et al.  Probabilistic Matrix Factorization , 2007, NIPS.

[42]  David Minnen,et al.  Full Resolution Image Compression with Recurrent Neural Networks , 2016, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).