Research map / Image generation / Autoregressive and latent generation
Energy-Based Models for Anomaly Detection: A Manifold Diffusion Recovery Approach
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 |
|---|---|---|---|
| Use MPDR-R instead of NAE (same autoencoder), MPDR training | reconstruction-energy EBM · CIFAR-10 vs SVHN | AUROC: 0.9352 -> 0.9807 | Helped |
| Increase single-autoencoder Dz from 32 to 128 (ablation) | MPDR-R single AE · CIFAR-10 vs SVHN | AUROC: 0.8271 -> 0.9886 | Helped |
| Increase Dz from 32 to 128 on CIFAR-10 vs Constant (ablation) | MPDR-R single AE · CIFAR-10 vs Constant | AUROC: 0.9877 -> 0.9651 | Hurt |
| Manifold ensemble (Dz 32,64,128) instead of single AE Dz=32 (ablation) | MPDR-R · CIFAR-10 vs SVHN | AUROC: 0.8271 -> 0.9807 | Helped |
| Manifold ensemble vs single AE Dz=128 on CIFAR-100 inlier vs CelebA (ablation) | MPDR-R · CIFAR-100 (ViT) vs CelebA | AUROC: 0.9526 -> 0.8662 | Hurt |
| Increase sigma from 0.01 to 0.3 (ablation) | MPDR-S · MNIST hold-out digit 9 | AUPR: 0.098 -> 0.558 | Helped |
| Increase sigma from 0.1 to 0.3 on EMNIST (ablation) | MPDR-S · MNIST vs EMNIST | AUPR: 0.995 -> 0.990 | Hurt |
| Increase Dz from 16 to 256 on Constant outliers (ablation) | MPDR-S · MNIST vs Constant | AUPR: 1.000 -> 0.866 | Hurt |
| Increase Dz from 16 to 128 on Omniglot (ablation) | MPDR-S · MNIST vs Omniglot | AUPR: 0.958 -> 0.999 | Helped |
| Increase gamma from 0.0001 to 0.1 (ablation) | MPDR-R · CIFAR-100 (ViT) vs SVHN | AUROC: 0.9932 -> 0.9424 | Hurt |
| Set gamma=0 instead of 0.0001 (ablation) | MPDR-R · CIFAR-100 (ViT) vs CelebA | AUROC: 0.8662 -> 0.8456 | Hurt |
| Apply MPDR on top of AE energy | AE · DCASE 2020 Toy Car | AUROC: 75.40 -> 81.54 | Helped |
Closest papers by meaning
- One-for-More: Continual Diffusion Model for Anomaly Detection (CVPR 2025 · similarity 0.85)
- A Unified Latent Schrodinger Bridge Diffusion Model for Unsupervised Anomaly Detection and Localization (CVPR 2025 · similarity 0.85)
- Distribution Prototype Diffusion Learning for Open-set Supervised Anomaly Detection (CVPR 2025 · similarity 0.85)
- On Diffusion Modeling for Anomaly Detection (ICLR 2024 · similarity 0.85)
- Diffusion Models Encode the Intrinsic Dimension of Data Manifolds (ICML 2024 · similarity 0.85)
- Learning Energy-Based Models by Cooperative Diffusion Recovery Likelihood (ICLR 2024 · similarity 0.85)
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.