Discrete distributions are learnable from metastable samples

TL;DR

This paper demonstrates that for discrete distributions, the true model can be recovered from samples generated by metastable states, even when these samples are far from the stationary distribution, by leveraging properties of conditional distributions.

Contribution

It provides a rigorous method to learn the true discrete distribution from metastable samples using conditional likelihood, applicable even with poor mixing in Markov chains.

Findings

Metastable samples' conditionals are close to the true distribution's conditionals.

Learning is possible despite large global divergence between metastable and true distributions.

Results extend to Ising models, enabling parameter and structure recovery.

Abstract

Physically motivated stochastic dynamics are often used to sample from high-dimensional distributions. However such dynamics often get stuck in specific regions of their state space and mix very slowly to the desired stationary state. This causes such systems to approximately sample from a metastable distribution which is usually quite different from the desired, stationary distribution of the dynamic. We rigorously show that, in the case of multi-variable discrete distributions, the true model describing the stationary distribution can be recovered from samples produced from a metastable distribution under minimal assumptions about the system. This follows from a fundamental observation that the single-variable conditionals of metastable distributions that satisfy a strong metastability condition are on average close to those of the stationary distribution. This holds even when the metastable distribution differs considerably from the true model in terms of global metrics like Kullback-Leibler divergence or total variation distance. This property allows us to learn the true model using a conditional likelihood based estimator, even when the samples come from a metastable distribution concentrated in a small region of the state space. Explicit examples of such metastable states can be constructed from regions that effectively bottleneck the probability flow and cause poor mixing of the Markov chain. For specific cases of binary pairwise undirected graphical models (i.e. Ising models), we extend our results to further rigorously show that data coming from metastable states can be used to learn the parameters of the energy function and recover the structure of the model.