Research map / Probabilistic and scientific modeling / Variational and generative sampling
No MCMC for me: Amortized sampling for fast and stable training of energy-based models
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 VERA instead of alternative unnormalized-model training methods | NICE trained on MNIST · MNIST | maximum likelihood: -2039 -> -1138 | Helped |
| remove entropy regularization from VERA (ablation) | NICE trained with VERA · MNIST | maximum likelihood: -1138 -> -1214 | Hurt |
| replace the variational importance-sampling entropy estimator with the HMC-based estimator (ablation) | VERA · MNIST | maximum likelihood: -1138 -> -1165 | Hurt |
| increase the classification weight from α=1 to α=100 (ablation) | VERA-trained JEM · CIFAR10 | classification accuracy: 76.1 -> 93.2 | Helped |
| increase the classification weight from α=1 to α=100 (ablation) | VERA-trained JEM · CIFAR100 | classification accuracy: 48.7 -> 72.2 | Helped |
| increase the classification weight from α=1 to α=100 (ablation) | VERA-trained JEM · CIFAR10 | FID: 27.5 -> 30.5 | Hurt |
| replace PCD training with VERA | JEM · CIFAR10 | FID: 38.4 -> 30.5 | Helped |
| replace PCD training with VERA | JEM · OOD detection across reported datasets | AUROC: 0.67 -> 0.83; 0.67 -> 0.73; 0.65 -> 0.86; 0.75 -> 0.33 | Mixed |
| replace VAT with VERA-trained JEM | semi-supervised classification · HEPMASS, CROP, HUMAN, and MNIST tabular | accuracy: 86.7 -> 88.3; 94.5 -> 94.9; 84.0 -> 89.9; 98.6 -> 98.6 | Mixed |
| use VERA instead of the official JEM implementation with default hyperparameters | JEM on CIFAR10 · CIFAR10 | training speed: 2.8x | Helped |
| replace PCD training with VERA | JEM training · CIFAR10 | training stability | Helped |
| remove entropy regularization (ablation) | VERA on StackedMNIST · StackedMNIST | number of modes captured: 999 -> 998 | No clear effect |
Closest papers by meaning
- Learning Energy-Based Model with Variational Auto-Encoder as Amortized Sampler (AAAI 2021 · similarity 0.91)
- Learning Energy-based Model via Dual-MCMC Teaching (NeurIPS 2023 · similarity 0.90)
- Joint Learning of Energy-based Models and their Partition Function (ICML 2025 · similarity 0.90)
- Energy-Based Modelling for Discrete and Mixed Data via Heat Equations on Structured Spaces (NeurIPS 2024 · similarity 0.89)
- Learning Energy-Based Models by Cooperative Diffusion Recovery Likelihood (ICLR 2024 · similarity 0.86)
- TabEBM: A Tabular Data Augmentation Method with Distinct Class-Specific Energy-Based Models (NeurIPS 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.