Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Learning Gaussian Processes by Minimizing PAC-Bayesian Generalization Bounds
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 |
|---|---|---|---|
| Train sparse GPs by minimizing the inverse-binary-KL PAC-Bayes bound instead of the VFE or FITC objective | Sparse GP regression · sarcos | generalization upper bound: by more than a factor of 3 | Helped |
| Train sparse GPs by minimizing the inverse-binary-KL PAC-Bayes bound instead of the VFE or FITC objective | Sparse GP regression · pol | generalization upper bound: by a factor of roughly 2 | Helped |
| Train sparse GPs by minimizing the inverse-binary-KL PAC-Bayes bound instead of the VFE or FITC objective | Sparse GP regression · kin40k | generalization upper bound: by a factor between 1.3 and 2 | Helped |
| Use the tighter inverse-binary-KL bound rather than the Pinsker-based square-root bound (ablation) | PAC-GP and PAC-SGP · Regression experiments | performance | Helped |
| Increase the number of inducing inputs (ablation) | kl-PAC-SGP · kin40k | generalization upper bound | Helped |
| Increase the number of inducing inputs (ablation) | kl-PAC-SGP · pol and sarcos | generalization upper bound | No clear effect |
| Increase the number of inducing inputs (ablation) | GP models · pol, sarcos, and kin40k | MSE | Helped |
| Replace the bounded 0-1 loss with the bounded distance-sensitive exponential loss (ablation) | PAC-SGP · pol | MSE: improvement w.r.t. MSE of 25% | Helped |
| Replace the bounded 0-1 loss with the bounded distance-sensitive exponential loss (ablation) | PAC-SGP · sarcos | MSE: little change | No clear effect |
| Replace the bounded 0-1 loss with the bounded distance-sensitive exponential loss (ablation) | PAC-SGP · kin40k | MSE: a decline of 12% | Hurt |
| Use ARD rather than a shared lengthscale (ablation) | Sparse GP regression · pol and kin40k | performance | Helped |
| Use VFE rather than FITC (ablation) | Sparse GP regression · pol, sarcos, and kin40k | MSE | Helped |
Closest papers by meaning
- New Bounds for Sparse Variational Gaussian Processes (ICML 2025 · similarity 0.90)
- Robust Gaussian Processes via Relevance Pursuit (NeurIPS 2024 · similarity 0.90)
- Robust and Conjugate Gaussian Process Regression (ICML 2024 · similarity 0.88)
- Turbocharging Gaussian Process Inference with Approximate Sketch-and-Project (NeurIPS 2025 · similarity 0.88)
- Sparse Gaussian Processes: Structured Approximations and Power-EP Revisited (NeurIPS 2025 · similarity 0.88)
- Computation-Aware Gaussian Processes: Model Selection And Linear-Time Inference (NeurIPS 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.