mlsanityData to 14 Oct 2026

Exploring bidirectional bounds for minimax-training of Energy-based models

arXiv 2506.04609 ↗First posted 5 Jun 2025cs.LG cs.CVcited by 0

Not on the map. Why?

Abstract (arXiv)

Energy-based models (EBMs) estimate unnormalized densities in an elegant framework, but they are generally difficult to train. Recent work has linked EBMs to generative adversarial networks, by noting that they can be trained through a minimax game using a variational lower bound. To avoid the instabilities caused by minimizing a lower bound, we propose to instead work with bidirectional bounds, meaning that we maximize a lower bound and minimize an upper bound when training the EBM. We investigate four different bounds on the log-likelihood derived from different perspectives. We derive lower bounds based on the singular values of the generator Jacobian and on mutual information. To upper bound the negative log-likelihood, we consider a gradient penalty-like bound, as well as one based on diffusion processes. In all cases, we provide algorithms for evaluating the bounds. We compare the different bounds to investigate, the pros and cons of the different approaches. Finally, we demonstrate that the use of bidirectional bounds stabilizes EBM training and yields high-quality density estimation and sample generation.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 51 of its 91 references that are in the map. First posted on arXiv 5 Jun 2025.

Builds on

51 of 91 references.

References · 51

  1. 2020Denoising Diffusion Probabilistic Modelscited by 21,076
  2. 2020Score-Based Generative Modeling through Stochastic Differential Equationscited by 9,122
  3. 2019Generative Modeling by Estimating Gradients of the Data Distributioncited by 4,101
  4. 2014Deep Learning Face Attributes in the Wildcited by 3,693
  5. 2016A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networkscited by 2,385
  6. 2017Improved Training of Wasserstein GANscited by 2,235
  7. 2015Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networkscited by 2,054
  8. 2023Consistency Modelscited by 1,822
  9. 2018Spectral Normalization for Generative Adversarial Networkscited by 1,636
  10. 2016Variational Inference: A Review for Statisticianscited by 1,472
  11. 2018Learning deep representations by mutual information estimation and maximizationcited by 1,273
  12. 2014NICE: Non-linear Independent Components Estimationcited by 1,058
38 more
  1. 2020Improved Techniques for Training Score-Based Generative Modelscited by 1,043
  2. 2018Deep Anomaly Detection with Outlier Exposurecited by 942
  3. 2021Maximum Likelihood Training of Score-Based Diffusion Modelscited by 672
  4. 2021Tackling the Generative Learning Trilemma with Denoising Diffusion GANscited by 516
  5. 2019Likelihood Ratios for Out-of-Distribution Detectioncited by 453
  6. 2017PixelCNN++: Improving the PixelCNN with Discretized Logistic Mixture Likelihood and Other Modificationscited by 417
  7. 2018Assessing Generative Models via Precision and Recallcited by 406
  8. 2021Knowledge Distillation in Iterative Generative Models for Improved Sampling Speedcited by 348
  9. 2019Learning Non-Convergent Non-Persistent Short-Run MCMC Toward Energy-Based Modelcited by 148
  10. 2020Learning Energy-Based Models by Diffusion Recovery Likelihoodcited by 123
  11. 2019On the Anatomy of MCMC-Based Maximum Likelihood Learning of Energy-Based Modelscited by 105
  12. 2021Soft Truncation: A Universal Training Technique of Score-based Diffusion Model for High Precision Score Estimationcited by 97
  13. 2020Your GAN is Secretly an Energy-based Model and You Should use Discriminator Driven Latent Samplingcited by 71
  14. 2020No MCMC for me: Amortized sampling for fast and stable training of energy-based modelscited by 64
  15. 2016Cooperative Training of Descriptor and Generator Networkscited by 60
  16. 2017Gradient Estimators for Implicit Modelscited by 58
  17. 2021Hierarchical VAEs Know What They Don't Knowcited by 58
  18. 2018Learning Descriptor Networks for 3D Shape Synthesis and Analysiscited by 56
  19. 2019Maximum Entropy Generators for Energy-Based Modelscited by 56
  20. 2018A Spectral Approach to Gradient Estimation for Implicit Distributionscited by 45
  21. 2020Learning Energy-Based Model with Variational Auto-Encoder as Amortized Samplercited by 43
  22. 2023Generative Modeling through the Semi-dual Formulation of Unbalanced Optimal Transportcited by 41
  23. 2018Uncertainty in the Variational Information Bottleneckcited by 39
  24. 2021The Geometry of Deep Generative Image Models and its Applicationscited by 35
  25. 2016Deep Directed Generative Models with Energy-Based Probability Estimationcited by 34
  26. 2018Divergence Triangle for Joint Training of Generator Model, Energy-based Model, and Inference Modelcited by 34
  27. 2020Joint Training of Variational Auto-Encoder and Latent Energy-Based Modelcited by 30
  28. 2019Prescribed Generative Adversarial Networkscited by 23
  29. 2020Regularized Autoencoders via Relaxed Injective Probability Flowcited by 23
  30. 2021A deep learning theory for neural networks grounded in physicscited by 23
  31. 2023Learning Energy-Based Models by Cooperative Diffusion Recovery Likelihoodcited by 19
  32. 2014Accurate and Conservative Estimates of MRF Log-likelihood using Reverse Annealingcited by 18
  33. 2023Guiding Energy-based Models via Contrastive Latent Variablescited by 18
  34. 2017Calibrating Energy-based Generative Adversarial Networkscited by 13
  35. 2016Generative Adversarial Networks as Variational Training of Energy Based Modelscited by 10
  36. 2021Bounds all around: training energy-based models with bidirectional boundscited by 10
  37. 2024Improving Adversarial Energy-Based Model via Diffusion Processcited by 7
  38. 2022Bi-level Doubly Variational Learning for Energy-based Latent Variable Modelscited by 6

50 most cited of 51 All 51 (JSON)

Built on by

Not cited yet.