Mixed variational flows for discrete variables

Gian C. Diluvi, Benjamin Bloem-Reddy, Trevor Campbell
Proceedings of The 27th International Conference on Artificial Intelligence and Statistics, PMLR 238:2431-2439, 2024.

Abstract

Variational flows allow practitioners to learn complex continuous distributions, but approximating discrete distributions remains a challenge. Current methodologies typically embed the discrete target in a continuous space—usually via continuous relaxation or dequantization—and then apply a continuous flow. These approaches involve a surrogate target that may not capture the original discrete target, might have biased or unstable gradients, and can create a difficult optimization problem. In this work, we develop a variational flow family for discrete distributions without any continuous embedding. First, we develop a measure-preserving and discrete (MAD) invertible map that leaves the discrete target invariant, and then create a mixed variational flow (MAD Mix) based on that map. Our family provides access to i.i.d. sampling and density evaluation with virtually no tuning effort. We also develop an extension to MAD Mix that handles joint discrete and continuous models. Our experiments suggest that MAD Mix produces more reliable approximations than continuous-embedding flows while requiring orders of magnitude less compute.

Cite this Paper


BibTeX
@InProceedings{pmlr-v238-diluvi24a, title = {Mixed variational flows for discrete variables}, author = {Diluvi, Gian C. and Bloem-Reddy, Benjamin and Campbell, Trevor}, booktitle = {Proceedings of The 27th International Conference on Artificial Intelligence and Statistics}, pages = {2431--2439}, year = {2024}, editor = {Dasgupta, Sanjoy and Mandt, Stephan and Li, Yingzhen}, volume = {238}, series = {Proceedings of Machine Learning Research}, month = {02--04 May}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v238/diluvi24a/diluvi24a.pdf}, url = {https://proceedings.mlr.press/v238/diluvi24a.html}, abstract = {Variational flows allow practitioners to learn complex continuous distributions, but approximating discrete distributions remains a challenge. Current methodologies typically embed the discrete target in a continuous space—usually via continuous relaxation or dequantization—and then apply a continuous flow. These approaches involve a surrogate target that may not capture the original discrete target, might have biased or unstable gradients, and can create a difficult optimization problem. In this work, we develop a variational flow family for discrete distributions without any continuous embedding. First, we develop a measure-preserving and discrete (MAD) invertible map that leaves the discrete target invariant, and then create a mixed variational flow (MAD Mix) based on that map. Our family provides access to i.i.d. sampling and density evaluation with virtually no tuning effort. We also develop an extension to MAD Mix that handles joint discrete and continuous models. Our experiments suggest that MAD Mix produces more reliable approximations than continuous-embedding flows while requiring orders of magnitude less compute.} }
Endnote
%0 Conference Paper %T Mixed variational flows for discrete variables %A Gian C. Diluvi %A Benjamin Bloem-Reddy %A Trevor Campbell %B Proceedings of The 27th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2024 %E Sanjoy Dasgupta %E Stephan Mandt %E Yingzhen Li %F pmlr-v238-diluvi24a %I PMLR %P 2431--2439 %U https://proceedings.mlr.press/v238/diluvi24a.html %V 238 %X Variational flows allow practitioners to learn complex continuous distributions, but approximating discrete distributions remains a challenge. Current methodologies typically embed the discrete target in a continuous space—usually via continuous relaxation or dequantization—and then apply a continuous flow. These approaches involve a surrogate target that may not capture the original discrete target, might have biased or unstable gradients, and can create a difficult optimization problem. In this work, we develop a variational flow family for discrete distributions without any continuous embedding. First, we develop a measure-preserving and discrete (MAD) invertible map that leaves the discrete target invariant, and then create a mixed variational flow (MAD Mix) based on that map. Our family provides access to i.i.d. sampling and density evaluation with virtually no tuning effort. We also develop an extension to MAD Mix that handles joint discrete and continuous models. Our experiments suggest that MAD Mix produces more reliable approximations than continuous-embedding flows while requiring orders of magnitude less compute.
APA
Diluvi, G.C., Bloem-Reddy, B. & Campbell, T.. (2024). Mixed variational flows for discrete variables. Proceedings of The 27th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 238:2431-2439 Available from https://proceedings.mlr.press/v238/diluvi24a.html.

Related Material