Research map / Optimization / Optimizers
Kernel Stein Discrepancy Descent
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 KSD Descent instead of SVGD | Bayesian logistic regression · 13 datasets | test accuracy | No clear effect |
| Use KSD Descent instead of SVGD | Bayesian independent component analysis · ICA, p in {2, 4, 8} | Amari distance | Hurt |
| Use KSD Descent instead of SVGD | KSD Descent · Standard 2D Gaussian toy example | distribution approximation | No clear effect |
| Use a high-temperature run with β = 0.1, then rerun at β = 1 | KSD Descent · Symmetric two-Gaussian mixture | convergence to target | Helped |
| Use annealing | KSD Descent · Bayesian independent component analysis | Amari distance | No clear effect |
| Initialize particles on the mixture's axis of symmetry | KSD Descent · Symmetric two-Gaussian mixture | convergence to target | Hurt |
| Initialize all particles on one side of the symmetry axis | KSD Descent · Symmetric two-Gaussian mixture | mixture-component coverage | Hurt |
| Use a step size that is too large | SVGD · 1D Gaussian toy example | convergence | Hurt |
| Use a step size that is too small | SVGD · 1D Gaussian toy example | convergence speed | Hurt |
Closest papers by meaning
- Sliced Kernelized Stein Discrepancy (ICLR 2021 · similarity 0.91)
- Gradient-Free Kernel Stein Discrepancy (NeurIPS 2023 · similarity 0.90)
- A Finite-Particle Convergence Rate for Stein Variational Gradient Descent (NeurIPS 2023 · similarity 0.89)
- Improved Finite-Particle Convergence Rates for Stein Variational Gradient Descent (ICLR 2025 · similarity 0.89)
- Stein Variational Gradient Descent as Moment Matching (NeurIPS 2018 · similarity 0.89)
- The Polynomial Stein Discrepancy for Assessing Moment Convergence (ICML 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.