Research map / Optimization / Optimizers
Efficient Learning of Generative Models via Finite-Difference Score Matching
Tianyu Pang, Kun Xu, Chongxuan LI, Yang Song, Stefano Ermon, Jun Zhu
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 |
|---|---|---|---|
| replace gradient-based SSM with finite-difference FD-SSM | SSM · deep EBM, MNIST | speedup: 2.9x | Improved |
| replace gradient-based DSM with finite-difference FD-DSM | DSM · deep EBM, MNIST | speedup: 1.7x | Improved |
| FD-DSM vs DSM final SM loss | DSM · deep EBM, MNIST | SM loss | Hurt |
| FD-SSM vs SSM on DKEF training time | SSM · DKEF Parkinsons | time per iteration: 110 ms -> 82.5 ms | Improved |
| FD-SSM vs SSM on DKEF test NLL | SSM · DKEF Parkinsons | NLL: 14.52 -> 13.69 | Improved |
| FD-SSM vs SSM on DKEF test NLL | SSM · DKEF RedWine | NLL: 13.34 -> 13.06 | Improved |
| FD-SSM vs SSM on DKEF test NLL | SSM · DKEF WhiteWine | NLL: 14.13 -> 14.10 | No clear effect |
| FD-SSM vs SSMVR on DKEF test NLL | SSMVR · DKEF Parkinsons | NLL: 13.26 -> 13.69 | Hurt |
| vary finite-difference step epsilon 0.1, 0.05, 0.02, 0.01 (ablation) | FD-SSM · DKEF Parkinsons | test NLL: 14.17, 13.51, 14.03, 14.00 | No clear effect |
| FD-SSMVR vs SSMVR for VAE implicit encoder training time | SSMVR · VAE MNIST | time: 5.04 ms -> 3.98 ms | Improved |
| FD-SSMVR vs SSMVR for VAE implicit encoder training time | SSMVR · VAE CelebA | time: 14.9 ms -> 9.38 ms | Improved |
| FD-SSMVR vs SSMVR for VAE sample quality | SSMVR · VAE CelebA | FID: 62.76 -> 64.85 | Hurt |
Closest papers by meaning
- Algorithm- and Data-Dependent Generalization Bounds for Diffusion Models (NeurIPS 2025 · similarity 0.86)
- Global Well-posedness and Convergence Analysis of Score-based Generative Models via Sharp Lipschitz Estimates (ICLR 2025 · similarity 0.86)
- Evaluating the design space of diffusion-based generative models (NeurIPS 2024 · similarity 0.85)
- Dimension-free Score Matching and Time Bootstrapping for Diffusion Models (NeurIPS 2025 · similarity 0.85)
- Score-based generative models are provably robust: an uncertainty quantification perspective (NeurIPS 2024 · similarity 0.85)
- Expressive Score-Based Priors for Distribution Matching with Geometry-Preserving Regularization (ICML 2025 · 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.
