Research map / Probabilistic and scientific modeling / Variational and generative sampling
Variational Inference with Gaussian Score Matching
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 GSM-VI instead of reparameterization BBVI | Gaussian variational inference · Synthetic and posteriorDB models | gradient evaluations to converge: 10-100x fewer gradient evaluations | Helped |
| Use GSM-VI instead of reparameterization BBVI | Gaussian variational inference · Gaussian targets with varying dimensions | convergence speed: 100 times more iterations | Helped |
| Use GSM-VI instead of reparameterization BBVI | Gaussian variational inference · 10-dimensional Gaussian targets with varying condition numbers | convergence robustness: c>100 | Helped |
| Use GSM-VI instead of reparameterization BBVI | Gaussian variational inference · Sinh-arcsinh targets | convergence speed: at least 10 times faster | Helped |
| Use GSM-VI instead of reparameterization BBVI | Gaussian variational inference · Sinh-arcsinh targets | average FKL | No clear effect |
| Apply GSM-VI to highly non-Gaussian targets | GSM-VI · Sinh-arcsinh targets (s ≥ 1 or |t-1| ≥ 0.8) | convergence stability: s ≥ 1 or |t-1| ≥ 0.8 | Hurt |
| Use batch size 2 instead of batch size 1 (ablation) | GSM-VI · Experiments across the paper | performance | Helped |
| Increase batch size beyond 2 (ablation) | GSM-VI · Experiments across the paper | performance | No clear effect |
| Vary the initialization of the variational distribution (ablation) | GSM-VI · posteriorDB models | qualitative conclusions | No clear effect |
| Use GSM-VI instead of reparameterization BBVI | Gaussian variational inference · posteriorDB models with Gaussian posteriors | solution stability | Helped |
| Use GSM-VI instead of reparameterization BBVI | Gaussian variational inference · posteriorDB models with non-Gaussian posteriors | variational approximation quality | No clear effect |
Closest papers by meaning
- Batch and match: black-box variational inference with a score-based divergence (ICML 2024 · similarity 0.93)
- Variational Inference with Mixtures of Isotropic Gaussians (NeurIPS 2025 · similarity 0.92)
- EigenVI: score-based variational inference with orthogonal function expansions (NeurIPS 2024 · similarity 0.92)
- Theoretical Guarantees for Variational Inference with Fixed-Variance Mixture of Gaussians (ICML 2024 · similarity 0.91)
- Least squares variational inference (NeurIPS 2025 · similarity 0.90)
- Generalized Variational Inference via Optimal Transport (AAAI 2024 · similarity 0.89)
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.