Hamiltonian Annealed Importance Sampling for partition function estimation
Machine Learning
2012-05-10 v1 Data Analysis, Statistics and Probability
Abstract
We introduce an extension to annealed importance sampling that uses Hamiltonian dynamics to rapidly estimate normalization constants. We demonstrate this method by computing log likelihoods in directed and undirected probabilistic image models. We compare the performance of linear generative models with both Gaussian and Laplace priors, product of experts models with Laplace and Student's t experts, the mc-RBM, and a bilinear generative model. We provide code to compare additional models.
Keywords
Cite
@article{arxiv.1205.1925,
title = {Hamiltonian Annealed Importance Sampling for partition function estimation},
author = {Jascha Sohl-Dickstein and Benjamin J. Culpepper},
journal= {arXiv preprint arXiv:1205.1925},
year = {2012}
}