EBMs Trained with Maximum Likelihood are Generator Models Trained with a Self-adverserial Loss
Preprint
Abstract
Maximum likelihood estimation is widely used in training Energy-based models (EBMs). Training requires samples from an unnormalized distribution, which is usually intractable, and in practice, these are obtained by MCMC algorithms such as Langevin dynamics. However, since MCMC in high-dimensional space converges extremely slowly, the current understanding of maximum likelihood training, which assumes approximate samples from the model can be drawn, is problematic. In this paper, we try to understand this training procedure by replacing Langevin dynamics with deterministic solutions of the associated gradient descent ODE. Doing so allows us to study the density induced by the dynamics (if the dynamics are invertible), and connect with GANs by treating the dynamics as generator models, the initial values as latent variables and the loss as optimizing a critic defined by the very same energy that determines the generator through its gradient. Hence the term - self-adversarial loss. We show that reintroducing the noise in the dynamics does not lead to a qualitative change in the behavior, and merely reduces the quality of the generator. We thus show that EBM training is effectively a self-adversarial procedure rather than maximum likelihood estimation.Keywords
All Related Versions
This publication has 3 references indexed in Scilit:
- Learning Energy-Based Models in High-Dimensional Spaces with Multiscale Denoising-Score MatchingEntropy, 2023
- Learning Energy-Based Model with Variational Auto-Encoder as Amortized SamplerProceedings of the AAAI Conference on Artificial Intelligence, 2021
- Generative VoxelNet: Learning Energy-Based Models for 3D Shape Synthesis and AnalysisIEEE Transactions on Pattern Analysis and Machine Intelligence, 2020