Meta-Learning MCMC Proposals

TL;DR

This paper introduces a meta-learning approach for constructing neural network-based MCMC proposals that generalize across models, improving inference efficiency and accuracy in diverse probabilistic tasks.

Contribution

It proposes a neural proposal parametrization for MCMC that generalizes across models, enabling fast, model-agnostic inference primitives without model-specific training.

Findings

Neural proposals outperform hand-tuned samplers in Gaussian mixture models.

Learned proposals achieve higher F1 scores than classical Gibbs sampling in NER.

Proposals generalize to structural motifs across different models.

Abstract

Effective implementations of sampling-based probabilistic inference often require manually constructed, model-specific proposals. Inspired by recent progresses in meta-learning for training learning agents that can generalize to unseen environments, we propose a meta-learning approach to building effective and generalizable MCMC proposals. We parametrize the proposal as a neural network to provide fast approximations to block Gibbs conditionals. The learned neural proposals generalize to occurrences of common structural motifs across different models, allowing for the construction of a library of learned inference primitives that can accelerate inference on unseen models with no model-specific training required. We explore several applications including open-universe Gaussian mixture models, in which our learned proposals outperform a hand-tuned sampler, and a real-world named entity recognition task, in which our sampler yields higher final F1 scores than classical single-site Gibbs sampling.