Research map / Learning theory and algorithms / Matrix and tensor methods
Optimal Preconditioning and Fisher Adaptive Langevin Sampling
Michalis Titsias
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 |
|---|---|---|---|
| learn a global preconditioner from score increments using the inverse empirical Fisher matrix | MALA · Caravan | Median ESS: 7.144 ± 0.663 -> 1920.903 ± 55.821 | Improved |
| learn a global preconditioner from score increments using the inverse empirical Fisher matrix | MALA · MNIST | Median ESS: 8.188 ± 0.399 -> 811.522 ± 19.165 | Improved |
| learn a global preconditioner from score increments using the inverse empirical Fisher matrix | MALA · Inhomogeneous Gaussian | Median ESS: 9.793 ± 0.655 -> 2002.579 ± 30.001 | Improved |
| learn a global preconditioner from score increments using the inverse empirical Fisher matrix | MALA · GP target | Median ESS: 2007.640 ± 104.867 -> 1923.753 ± 95.820 | Mixed |
| use score-function differences without Rao-Blackwellization (ablation) | FisherMALA · GP target | Median ESS: 1923.753 ± 95.820 -> 1916.990 ± 85.208 | No clear effect |
| use paired mean and covariance stochastic estimation instead of score increments (ablation) | FisherMALA · GP target | Median ESS: 1923.753 ± 95.820 -> 1583.570 ± 109.241 | Hurt |
| use paired mean and covariance stochastic estimation instead of score increments (ablation) | FisherMALA · MNIST | Median ESS: 811.522 ± 19.165 -> 13.819 ± 1.127 | Hurt |
| adapt directly from score vectors rather than centered score increments (ablation) | FisherMALA · Caravan | Median ESS: 1920.903 ± 55.821 -> 14.226 ± 4.429 | Hurt |
| adapt directly from score vectors rather than centered score increments (ablation) | FisherMALA · GP target | Median ESS: 1923.753 ± 95.820 -> 518.924 ± 579.639 | Hurt |
Closest papers by meaning
- Preconditioned Langevin Dynamics with Score-based Generative Models for Infinite-Dimensional Linear Bayesian Inverse Problems (NeurIPS 2025 · similarity 0.91)
- Preconditioned Stochastic Gradient Langevin Dynamics for Deep Neural Networks (AAAI 2016 · similarity 0.89)
- The Expected Loss of Preconditioned Langevin Dynamics Reveals the Hessian Rank (AAAI 2024 · similarity 0.89)
- Dimension-Uniform Discretization Analysis of Preconditioned Annealed Langevin Dynamics for Multimodal Gaussian Mixtures (NeurIPS 2026 · similarity 0.87)
- Sample as you Infer: Predictive Coding with Langevin Dynamics (ICML 2024 · similarity 0.87)
- From stability of Langevin diffusion to convergence of proximal MCMC for non-log-concave sampling (NeurIPS 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.
