Bounds all around: training energy-based models with bidirectional bounds

Abstract

Energy-based models (EBMs) provide an elegant framework for density estimation, but they are notoriously difficult to train. Recent work has established links to generative adversarial networks, where the EBM is trained through a minimax game with a variational value function. We propose a bidirectional bound on the EBM log-likelihood, such that we maximize a lower bound and minimize an upper bound when solving the minimax game. We link one bound to a gradient penalty that stabilizes training, thereby providing grounding for best engineering practice. To evaluate the bounds we develop a new and efficient estimator of the Jacobi-determinant of the EBM generator. We demonstrate that these developments significantly stabilize training and yield high-quality density estimation and sample generation.

This paper has been accepted by NeurIPS 2021

References in corpus (7)

• Deep Anomaly Detection with Outlier Exposure
• Improving Generalization and Stability of Generative Adversarial Networks
• Maximum Entropy Generators for Energy-Based Models
• Calibrating Energy-based Generative Adversarial Networks
• The Geometry of Deep Generative Image Models and its Applications
• Generative Adversarial Networks as Variational Training of Energy Based Models
• A deep learning theory for neural networks grounded in physics