Hyper-parameter Learning for Sparse Structured Probabilistic Models

Tatiana Shpakova, Francis Bach, Mike Davies

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

In this paper, we consider the estimation of hyperparameters for regularization terms commonly used for obtaining structured sparse parameters in signal estimation problems, such as signal denoising. By considering the convex regularization terms as negative log-densities, we propose approximate maximum likelihood estimation for estimating parameters for continuous log-supermodular distributions, which is a key property that many sparse priors have. We then show how »perturb-and-MAP» ideas based on the Gumbel distribution and efficient discretization can be used to approximate the log-partition function for these models, which is a crucial step for approximate maximum likelihood estimation. We illustrate our estimation procedure on a set of experiments with flow-based priors and signal denoising.

Original languageEnglish
Title of host publication2019 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2019 - Proceedings
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages3347-3351
Number of pages5
ISBN (Electronic)9781479981311
DOIs
Publication statusPublished - May 2019
Event44th IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2019 - Brighton, United Kingdom
Duration: 12 May 201917 May 2019

Publication series

NameICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings
Volume2019-May
ISSN (Print)1520-6149

Conference

Conference44th IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2019
CountryUnited Kingdom
CityBrighton
Period12/05/1917/05/19

Keywords

  • hyperparameter learning
  • maximum likelihood
  • sparsity-inducing norms

Fingerprint Dive into the research topics of 'Hyper-parameter Learning for Sparse Structured Probabilistic Models'. Together they form a unique fingerprint.

Cite this