Hamiltonian Monte Carlo with categorical parameters using the Concrete distribution
(2024) Sixth Symposium on Advances in Approximate Bayesian Inference- Abstract
- We introduce a method to enable Hamiltonian Monte Carlo (HMC) to simulate from mixed continuous and discrete posterior distributions. In particular, we show how the "Gumbel Max Trick" and the Concrete (Gumbel-softmax) distribution can be used for constructing a continuous approximation of a categorical distribution, and how this distribution can be efficiently implemented for HMC. We also illustrate how the Concrete distribution can be incorporated into a latent discrete parameter model, resulting in the Concrete Mixture model.
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/record/57bbfa82-3a66-4442-b8e8-c14f6f5b63cf
- author
- Torgander, Jakob ; Magnusson, Måns and Wallin, Jonas LU
- organization
- publishing date
- 2024-07-21
- type
- Contribution to conference
- publication status
- published
- subject
- keywords
- Hamiltonian Monte Carlo, MCMC, robabilistic machine learning, Gumbel max trick
- pages
- 12 pages
- conference name
- Sixth Symposium on Advances in Approximate Bayesian Inference
- conference location
- Vienna, Austria
- conference dates
- 2024-07-21 - 2024-07-21
- language
- English
- LU publication?
- yes
- id
- 57bbfa82-3a66-4442-b8e8-c14f6f5b63cf
- alternative location
- https://openreview.net/forum?id=vyIUhU5LUr
- date added to LUP
- 2025-03-05 10:00:19
- date last changed
- 2025-06-23 10:39:42
@misc{57bbfa82-3a66-4442-b8e8-c14f6f5b63cf, abstract = {{We introduce a method to enable Hamiltonian Monte Carlo (HMC) to simulate from mixed continuous and discrete posterior distributions. In particular, we show how the "Gumbel Max Trick" and the Concrete (Gumbel-softmax) distribution can be used for constructing a continuous approximation of a categorical distribution, and how this distribution can be efficiently implemented for HMC. We also illustrate how the Concrete distribution can be incorporated into a latent discrete parameter model, resulting in the Concrete Mixture model.}}, author = {{Torgander, Jakob and Magnusson, Måns and Wallin, Jonas}}, keywords = {{Hamiltonian Monte Carlo; MCMC; robabilistic machine learning; Gumbel max trick}}, language = {{eng}}, month = {{07}}, title = {{Hamiltonian Monte Carlo with categorical parameters using the Concrete distribution}}, url = {{https://openreview.net/forum?id=vyIUhU5LUr}}, year = {{2024}}, }