Research map / Probabilistic and scientific modeling / Variational and generative sampling
Particle-based Variational Inference with Preconditioned Functional Gradient Flow
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 PFG instead of SVGD or SGLD | Hierarchical logistic regression · covtype and german | accuracy: covtype: 75.1 -> 75.7; german: 65.1 -> 67.6 | Helped |
| Use PFG instead of SVGD or SGLD | Hierarchical logistic regression · covtype and german | nll: covtype: 0.58 -> 0.51; german: 0.71 -> 0.64 | Helped |
| Use PFG instead of M-SVGD or pSGLD | Bayesian neural networks · Boston | test RMSE: 2.72$_{\pm0.17}$ -> 2.47$_{\pm0.11}$ | Helped |
| Use PFG instead of M-SVGD or pSGLD | Bayesian neural networks · Energy | test RMSE: 0.89$_{\pm0.10}$ -> 0.48$_{\pm0.04}$ | Helped |
| Use PFG instead of SVGD | Particle-based inference · Sonar, 1,000 particles | time: 19.10$_{\pm0.52}$ -> 11.56$_{\pm0.24}$ | Helped |
| Use PFG instead of SVGD | Particle-based inference · Sonar, 5 particles | time: 5.20$_{\pm0.10}$ -> 7.87$_{\pm0.16}$ | Hurt |
| Use PFG instead of JKO methods | Bayesian logistic regression · Covtype | accuracy: JKO-ICNN: 75.0 -> 75.7; JKO-Variational: 75.3 -> 75.7 | Helped |
| Use diagonal preconditioning rather than identity preconditioning (ablation) | PFG · ill-conditioned Gaussian sampling | KL divergence | Helped |
| Use Hessian-aware preconditioning rather than identity (ablation) | PFG · Gaussian sampling | KL convergence bound: (1-σ²_d/(2σ²_1))^(2t) C_0 -> 1/4^t C_0 | Helped |
| Use a neural-network function class rather than an RBF kernel function class (ablation) | Particle-based VI · 10-cluster Gaussian mixture | sampling quality | Helped |
| Use an RBF function class rather than a linear function class (ablation) | SVGD · 20-dim N(0,I) variance fitting | variance: 1.01 -> 0.35 | Hurt |
| Use a neural-network function class rather than a linear function class (ablation) | PFG · 20-dim N(0,I) variance fitting | variance: 1.00 -> 0.99 | No clear effect |
Closest papers by meaning
- Particle-based Variational Inference with Generalized Wasserstein Gradient Flow (NeurIPS 2023 · similarity 0.94)
- Functional Gradient Flows for Constrained Sampling (NeurIPS 2024 · similarity 0.93)
- Approximate Bayesian Inference with Stein Functional Variational Gradient Descent (ICLR 2023 · similarity 0.92)
- Particle Semi-Implicit Variational Inference (NeurIPS 2024 · similarity 0.90)
- Globally Convergent Variational Inference (NeurIPS 2024 · similarity 0.89)
- Theoretical Guarantees for Variational Inference with Fixed-Variance Mixture of Gaussians (ICML 2024 · similarity 0.88)
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.