machine learning

Learning Latent Energy-Based Models via Interacting Particle Langevin Dynamics

arXiv:2510.12311

summary

The paper proposes a continuous‑time framework using interacting particle Langevin dynamics to learn latent variable models with energy‑based priors, provides a discretized algorithm with convergence guarantees, and demonstrates its efficiency on synthetic and image data.

Abstract

We develop interacting particle algorithms for learning latent variable models with energy-based priors. To do so, we leverage recent developments in particle-based methods for solving maximum marginal likelihood estimation (MMLE) problems. Specifically, we provide a continuous-time framework for learning latent energy-based models, by defining stochastic differential equations (SDEs) that provably solve the MMLE problem. We obtain a practical algorithm as a discretisation of these SDEs and provide theoretical guarantees for the convergence of the proposed algorithm. Finally, we empirically validate the effectiveness of our method on synthetic and image datasets and demonstrate that using a particle based approach offers significant improvement in computational efficiency.

Accepted as an oral presentation at UAI'26

Topics & keywords

#latent variable models#energy-based models#interacting particle methods#langevin dynamics#stochastic differential equationslatent energy-based modelsinteracting particle Langevin dynamicsmaximum marginal likelihoodSDE discretizationconvergence guarantees
Learning Latent Energy-Based Models via Interacting Particle Langevin Dynamics · wovepaper