Research map / Probabilistic and scientific modeling / Variational and generative sampling
Particles Don’t Care About Z: Towards Scaling Entropy Estimation of Unnormalized Densities
Safa Messaoud, Skander Charni, Elaa Bouazza, Ali Pourghasemi Fatideh, Halima Bensmail
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| restore the trace-of-Hessian correction term in the entropy estimate (ablation) | MET-SVGD · CIFAR-10 EBM training | training stability | Improved |
| add the step-size bound for global invertibility (ablation) | MET-SVGD · CIFAR-10 EBM training | training stability | Improved |
| replace median-heuristic bandwidth with learned bandwidth (ablation) | MET-SVGD · CIFAR-10 EBM | FID and stability | Improved |
| use adaptive number of steps via Stein discrepancy (ablation) | MET-SVGD · CIFAR-10 EBM | training stability: 64x improvement over P-SVGD | Improved |
| add Metropolis-Hastings step with contrastive loss (ablation) | MET-SVGD · CIFAR-10 EBM | training stability | Hurt |
| add Metropolis-Hastings step to handle non-smooth targets (ablation) | MET-SVGD · Moon dataset EBM | accuracy | Improved |
| learn kernel bandwidth (ablation) | MET-SVGD · MaxEnt RL | return | Improved |
| learn step size without the correction term (ablation) | MET-SVGD · MaxEnt RL | return | Hurt |
| use adaptive number of SVGD steps (ablation) | MET-SVGD · MaxEnt RL | return | Hurt |
| add MH step with epsilon-greedy schedule and increase particles from 10 to 64 (ablation) | MET-SVGD · MaxEnt RL | return | Improved |
| restore trace-of-Hessian term (ablation) | MET-SVGD · High-dimensional Gaussian entropy | entropy accuracy | Improved |
Closest papers by meaning
- Provably Fast Finite Particle Variants of SVGD via Virtual Particle Stochastic Approximation (NeurIPS 2023 · similarity 0.90)
- Towards Understanding the Dynamics of Gaussian-Stein Variational Gradient Descent (NeurIPS 2023 · similarity 0.89)
- Understanding the Variance Collapse of SVGD in High Dimensions (ICLR 2022 · similarity 0.88)
- ELBOing Stein: Variational Bayes with Stein Mixture Inference (ICLR 2025 · similarity 0.88)
- Theoretical Guarantees for Variational Inference with Fixed-Variance Mixture of Gaussians (ICML 2024 · similarity 0.87)
- Improved Finite-Particle Convergence Rates for Stein Variational Gradient Descent (ICLR 2025 · similarity 0.87)
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.
