LSB: Local Self-Balancing MCMC in Discrete Spaces

Emanuele Sansone
Proceedings of the 39th International Conference on Machine Learning, PMLR 162:19205-19220, 2022.

Abstract

We present the Local Self-Balancing sampler (LSB), a local Markov Chain Monte Carlo (MCMC) method for sampling in purely discrete domains, which is able to autonomously adapt to the target distribution and to reduce the number of target evaluations required to converge. LSB is based on (i) a parametrization of locally balanced proposals, (ii) an objective function based on mutual information and (iii) a self-balancing learning procedure, which minimises the proposed objective to update the proposal parameters. Experiments on energy-based models and Markov networks show that LSB converges using a smaller number of queries to the oracle distribution compared to recent local MCMC samplers.

Cite this Paper

BibTeX
@InProceedings{pmlr-v162-sansone22a, title = {{LSB}: Local Self-Balancing {MCMC} in Discrete Spaces}, author = {Sansone, Emanuele}, booktitle = {Proceedings of the 39th International Conference on Machine Learning}, pages = {19205--19220}, year = {2022}, editor = {Chaudhuri, Kamalika and Jegelka, Stefanie and Song, Le and Szepesvari, Csaba and Niu, Gang and Sabato, Sivan}, volume = {162}, series = {Proceedings of Machine Learning Research}, month = {17--23 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v162/sansone22a/sansone22a.pdf}, url = {https://proceedings.mlr.press/v162/sansone22a.html}, abstract = {We present the Local Self-Balancing sampler (LSB), a local Markov Chain Monte Carlo (MCMC) method for sampling in purely discrete domains, which is able to autonomously adapt to the target distribution and to reduce the number of target evaluations required to converge. LSB is based on (i) a parametrization of locally balanced proposals, (ii) an objective function based on mutual information and (iii) a self-balancing learning procedure, which minimises the proposed objective to update the proposal parameters. Experiments on energy-based models and Markov networks show that LSB converges using a smaller number of queries to the oracle distribution compared to recent local MCMC samplers.} }
Endnote
%0 Conference Paper %T LSB: Local Self-Balancing MCMC in Discrete Spaces %A Emanuele Sansone %B Proceedings of the 39th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2022 %E Kamalika Chaudhuri %E Stefanie Jegelka %E Le Song %E Csaba Szepesvari %E Gang Niu %E Sivan Sabato %F pmlr-v162-sansone22a %I PMLR %P 19205--19220 %U https://proceedings.mlr.press/v162/sansone22a.html %V 162 %X We present the Local Self-Balancing sampler (LSB), a local Markov Chain Monte Carlo (MCMC) method for sampling in purely discrete domains, which is able to autonomously adapt to the target distribution and to reduce the number of target evaluations required to converge. LSB is based on (i) a parametrization of locally balanced proposals, (ii) an objective function based on mutual information and (iii) a self-balancing learning procedure, which minimises the proposed objective to update the proposal parameters. Experiments on energy-based models and Markov networks show that LSB converges using a smaller number of queries to the oracle distribution compared to recent local MCMC samplers.
APA
Sansone, E.. (2022). LSB: Local Self-Balancing MCMC in Discrete Spaces. Proceedings of the 39th International Conference on Machine Learning, in Proceedings of Machine Learning Research 162:19205-19220 Available from https://proceedings.mlr.press/v162/sansone22a.html.

Related Material