Persistently Trained, Diffusion-assisted Energy-based Models

Xinwei Zhang, Zhiqiang Tan, Zhijian Ou
4/21/2023
stat.MLcs.LG

No official code found. Best implementation: XinweiZhang/DAEBM (4 stars, last commit 2026-01-26).

Abstract

Maximum likelihood (ML) learning for energy-based models (EBMs) is challenging, partly due to non-convergence of Markov chain Monte Carlo.Several variations of ML learning have been proposed, but existing methods all fail to achieve both post-training image generation and proper density estimation. We propose to introduce diffusion data and learn a joint EBM, called diffusion assisted-EBMs, through persistent training (i.e., using persistent contrastive divergence) with an enhanced sampling algorithm to properly sample from complex, multimodal distributions. We present results from a 2D illustrative experiment and image experiments and demonstrate that, for the first time for image data, persistently trained EBMs can {\it simultaneously} achieve long-run stability, post-training image generation, and superior out-of-distribution detection.

View on arXivView PDF

Code Implementations(1)

XinweiZhang/DAEBMCommunity100%
40R, ShellNov 19, 20228 months ago

Similar Papers

Learning Energy-Based Models by Cooperative Diffusion Recovery Likelihood

88% match1 repo72023

Learning Energy-Based Models by Cooperative Diffusion Recovery Likelihood

88% match1 repo72026

Training Energy-Based Models with Diffusion Contrastive Divergences

88% match2023

Contrastive Energy Prediction for Exact Energy-Guided Diffusion Sampling in Offline Reinforcement Learning

84% match1 repo362023

Exploring Continual Learning of Diffusion Models

83% match2023

Maximum Entropy Inverse Reinforcement Learning of Diffusion Models with Energy-Based Models

83% match1 repo332024

Cite this paper

@article{zhang2023persistently,
  title  = {Persistently Trained, Diffusion-assisted Energy-based Models},
  author = {Xinwei Zhang and Zhiqiang Tan and Zhijian Ou},
  year   = {2023},
  eprint = {2304.10707},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2304.10707v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2304.10707.svg)](https://codeofpaper.com/paper/2304.10707)

Discussion