Research map / Probabilistic and scientific modeling / Variational and generative sampling
Entropy-based Training Methods for Scalable Neural Implicit Samplers
Weijian Luo, Boya Zhang, Zhihua Zhang
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 |
|---|---|---|---|
| Use KL-trained one-step neural sampler instead of LD/HMC with 500 iterations | KL-NS · 2D targets | inference time (s), MOG2: 0.2168 -> 0.0014 (vs LD); 1.7125 -> 0.0014 (vs HMC) | Improved |
| Use KL-NS instead of Fisher-NS | Neural sampler · Funnel | KSD: 7.837 -> 0.115 | Improved |
| Use KL-NS instead of KSD-NS | Neural sampler · Rosenbrock | KSD: 1.531 -> 0.123 | Improved |
| Use KL-NS instead of Coupling Flow | Neural sampler · Rosenbrock | KSD: 0.150 -> 0.123 | Improved |
| Use SVGD instead of neural samplers | Sampler · 2D targets | KSD: 0.044 -> 0.104 (MOG2, SVGD vs KL-NS) | Improved |
| Use KL-NS instead of LD with 500 iterations | KL-NS · 2D targets | KSD, MOG2: 0.099 -> 0.104 | No clear effect |
| Use Fisher-NS instead of SGLD, DSVI, SVGD and SteinGAN | Fisher-NS · Covertype | test accuracy (%): 75.37% -> 76.22% | Improved |
| Use KL-IS instead of Fisher-NS (ablation) | Implicit sampler · Covertype | test accuracy (%): 76.22% -> 75.95% | Hurt |
| Replace annealed Langevin EBM sampling (250 NFE) with one-NFE KL sampler | KL Sampler · MNIST EBM | FID: 20.95 -> 22.29 | Mixed |
| Replace EBM sampling at 100 NFE with one-NFE KL sampler | KL Sampler · MNIST EBM | FID: 30.35 -> 22.29 | Improved |
| Reduce EBM NFE from 250 to 50 (ablation) | EBM annealed Langevin · MNIST EBM | FID: 20.95 -> 52.55 | Hurt |
| Reduce EBM NFE from 150 to 100 (ablation) | EBM annealed Langevin · MNIST EBM | FID: 21.32 -> 30.35 | Hurt |
Closest papers by meaning
- Swift Sampler: Efficient Learning of Sampler by 10 Parameters (NeurIPS 2024 · similarity 0.87)
- Scalable Discrete Diffusion Samplers: Combinatorial Optimization and Statistical Physics (ICLR 2025 · similarity 0.87)
- Discrete Neural Flow Samplers with Locally Equivariant Transformer (NeurIPS 2025 · similarity 0.86)
- MDNS: Masked Diffusion Neural Sampler via Stochastic Optimal Control (NeurIPS 2025 · similarity 0.86)
- LEAPS: A discrete neural sampler via locally equivariant networks (ICML 2025 · similarity 0.86)
- Accelerating Neural Field Training via Soft Mining (CVPR 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.
