Research map / Learning theory and algorithms / Sample complexity
Bounds all around: training energy-based models with bidirectional bounds
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace the bidirectional upper bound with a zero-centered gradient penalty (ablation) | EBM-BB · CIFAR-10 | FID: 28.63 ± 0.290 -> 35.42 ± 0.582 | Hurt |
| Replace the bidirectional upper bound with a zero-centered gradient penalty (ablation) | EBM-BB · ANIMEFACE | FID: 12.75 ± 0.045 -> 20.53 ± 0.524 | Hurt |
| Replace the bidirectional upper bound with a zero-centered gradient penalty (ablation) | EBM-BB · ANIMEFACE | Inception Score: 2.26 ± 0.005 -> 2.26 ± 0.017 | No clear effect |
| Replace the bidirectional upper bound with a zero-centered gradient penalty (ablation) | EBM-BB · CIFAR-10 | Anisotropy index: 0.9431 ± 0.0191 -> 1.0688 ± 0.0226 | Hurt |
| Replace the bidirectional upper bound with a zero-centered gradient penalty (ablation) | EBM-BB · ANIMEFACE | Anisotropy index: 1.8016 ± 0.1345 -> 3.4560 ± 0.0521 | Hurt |
| Replace the bidirectional upper bound with a zero-centered gradient penalty (ablation) | EBM-BB · CIFAR-10 / SVHN | AUROC: 0.88 -> 0.66 | Hurt |
| Replace the bidirectional upper bound with a zero-centered gradient penalty (ablation) | EBM-BB · CIFAR-10 / CIFAR-100 | AUROC: 0.53 -> 0.64 | Helped |
| Replace the bidirectional upper bound with a zero-centered gradient penalty (ablation) | EBM-BB · ANIMEFACE / Bedroom | AUROC: 0.53 -> 0.67 | Helped |
| Use Hutchinson's entropy estimator instead of the proposed estimator (ablation) | EBM-BB entropy estimation · MNIST | Entropy estimate | Mixed |
| Remove batch normalization from the energy function (ablation) | EBM-BB · ANIMEFACE | Training stability | Helped |
| Replace latent dimension normalization with the squared norm of a random vector in the upper-bound coefficient (ablation) | EBM-BB · High-dimensional data | Generation quality | Helped |
Closest papers by meaning
- Improving Adversarial Energy-Based Model via Diffusion Process (ICML 2024 · similarity 0.90)
- Calibrating Energy-based Generative Adversarial Networks (ICLR 2017 · similarity 0.89)
- Energy-Inspired Models: Learning with Sampler-Induced Distributions (NeurIPS 2019 · similarity 0.88)
- Scalable Energy-Based Models via Adversarial Training: Unifying Discrimination and Generation (ICLR 2026 · similarity 0.88)
- Learning Energy-Based Model with Variational Auto-Encoder as Amortized Sampler (AAAI 2021 · similarity 0.87)
- Learning Energy-Based Models by Cooperative Diffusion Recovery Likelihood (ICLR 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.