Research map / Optimization / Optimizers
A Stein variational Newton method
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 |
|---|---|---|---|
| add a Newton-like second-order update instead of gradient descent | SVGD · Two-dimensional double banana | convergence speed | Helped |
| add a Newton-like second-order update instead of gradient descent | SVGD · 100-dimensional conditioned diffusion | posterior reconstruction | Helped |
| use a scaled Hessian kernel instead of an isotropic kernel (ablation) | SVN · Discretization-invariant inverse problem, d=40, 60, 80, 100 | trace posterior covariance: SVN-H: 0.1271, 0.1281, 0.1304, 0.1293; SVN-I: 0.0925, 0.0925, 0.0925, 0.0923 | Helped |
| use a scaled Hessian kernel instead of an isotropic kernel (ablation) | SVN · Non-discretization-invariant inverse problem, d=40, 60, 80, 100 | trace posterior covariance: SVN-H: 37.7331, 55.8354, 73.6383, 90.7689; SVN-I: 8.7133, 8.2588, 7.9862, 7.6876 | Helped |
| use a scaled Hessian kernel instead of an isotropic kernel (ablation) | SVN · Non-discretization-invariant inverse problem, d=40, 60, 80, 100 | average posterior mean: SVN-H: 0.0037, 0.0025, 0.0019, 0.0015; SVN-I: 0.0037, 0.0025, 0.0019, 0.0015 | No clear effect |
| use a scaled Hessian kernel instead of an isotropic kernel (ablation) | SVN · Discretization-invariant inverse problem, d=40, 60, 80, 100 | average posterior mean: SVN-H: 0.4658, 0.4634, 0.4623, 0.4614; SVN-I: 0.4657, 0.4633, 0.4622, 0.4615 | No clear effect |
| use the isotropic kernel instead of the scaled Hessian kernel (ablation) | SVN · 100-dimensional conditioned diffusion | credible interval accuracy | Hurt |
| use the block-diagonal or Newton-CG approximation instead of the full Newton system (ablation) | SVN · Double banana and nonlinear regression | sample quality | No clear effect |
| use SVN-H instead of SVGD-I | Bayesian neural network · Yacht hydrodynamics validation set | prediction error distribution | Helped |
Closest papers by meaning
- Stein Variational Gradient Descent With Matrix-Valued Kernels (NeurIPS 2019 · similarity 0.94)
- Stein Variational Gradient Descent as Moment Matching (NeurIPS 2018 · similarity 0.94)
- A Non-Asymptotic Analysis for Stein Variational Gradient Descent (NeurIPS 2020 · similarity 0.93)
- Stein Variational Gradient Descent Without Gradient (ICML 2018 · similarity 0.93)
- ELBOing Stein: Variational Bayes with Stein Mixture Inference (ICLR 2025 · similarity 0.90)
- Improved Finite-Particle Convergence Rates for Stein Variational Gradient Descent (ICLR 2025 · 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.