Source Separation with Deep Generative Priors

Vivek Jayaram, John Thickstun
Proceedings of the 37th International Conference on Machine Learning, PMLR 119:4724-4735, 2020.

Abstract

Despite substantial progress in signal source separation, results for richly structured data continue to contain perceptible artifacts. In contrast, recent deep generative models can produce authentic samples in a variety of domains that are indistinguishable from samples of the data distribution. This paper introduces a Bayesian approach to source separation that uses deep generative models as priors over the components of a mixture of sources, and noise-annealed Langevin dynamics to sample from the posterior distribution of sources given a mixture. This decouples the source separation problem from generative modeling, enabling us to directly use cutting-edge generative models as priors. The method achieves state-of-the-art performance for MNIST digit separation. We introduce new methodology for evaluating separation quality on richer datasets, providing quantitative evaluation and qualitative discussion of results for CIFAR-10 image separation. We also provide qualitative results on LSUN.

Cite this Paper


BibTeX
@InProceedings{pmlr-v119-jayaram20a, title = {Source Separation with Deep Generative Priors}, author = {Jayaram, Vivek and Thickstun, John}, booktitle = {Proceedings of the 37th International Conference on Machine Learning}, pages = {4724--4735}, year = {2020}, editor = {Hal Daumé III and Aarti Singh}, volume = {119}, series = {Proceedings of Machine Learning Research}, month = {13--18 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v119/jayaram20a/jayaram20a.pdf}, url = { http://proceedings.mlr.press/v119/jayaram20a.html }, abstract = {Despite substantial progress in signal source separation, results for richly structured data continue to contain perceptible artifacts. In contrast, recent deep generative models can produce authentic samples in a variety of domains that are indistinguishable from samples of the data distribution. This paper introduces a Bayesian approach to source separation that uses deep generative models as priors over the components of a mixture of sources, and noise-annealed Langevin dynamics to sample from the posterior distribution of sources given a mixture. This decouples the source separation problem from generative modeling, enabling us to directly use cutting-edge generative models as priors. The method achieves state-of-the-art performance for MNIST digit separation. We introduce new methodology for evaluating separation quality on richer datasets, providing quantitative evaluation and qualitative discussion of results for CIFAR-10 image separation. We also provide qualitative results on LSUN.} }
Endnote
%0 Conference Paper %T Source Separation with Deep Generative Priors %A Vivek Jayaram %A John Thickstun %B Proceedings of the 37th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2020 %E Hal Daumé III %E Aarti Singh %F pmlr-v119-jayaram20a %I PMLR %P 4724--4735 %U http://proceedings.mlr.press/v119/jayaram20a.html %V 119 %X Despite substantial progress in signal source separation, results for richly structured data continue to contain perceptible artifacts. In contrast, recent deep generative models can produce authentic samples in a variety of domains that are indistinguishable from samples of the data distribution. This paper introduces a Bayesian approach to source separation that uses deep generative models as priors over the components of a mixture of sources, and noise-annealed Langevin dynamics to sample from the posterior distribution of sources given a mixture. This decouples the source separation problem from generative modeling, enabling us to directly use cutting-edge generative models as priors. The method achieves state-of-the-art performance for MNIST digit separation. We introduce new methodology for evaluating separation quality on richer datasets, providing quantitative evaluation and qualitative discussion of results for CIFAR-10 image separation. We also provide qualitative results on LSUN.
APA
Jayaram, V. & Thickstun, J.. (2020). Source Separation with Deep Generative Priors. Proceedings of the 37th International Conference on Machine Learning, in Proceedings of Machine Learning Research 119:4724-4735 Available from http://proceedings.mlr.press/v119/jayaram20a.html .

Related Material