Research map / Learning theory and algorithms / Kernels and regression theory
Certifying Deep Network Risks and Individual Predictions with PAC-Bayes Loss via Localized Priors
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes localized PAC-Bayes priors to produce tighter, practical generalization certificates for deep networks, covering both risk bounds and individual predictions.
Problem
PAC-Bayes generalization certificates for deep networks are rarely practical: simple theoretical priors give vacuous bounds, while data-dependent priors are needed for tight guarantees.
Approach
Uses localized priors within the PAC-Bayes framework to obtain non-vacuous certificates on deep network risk and on individual predictions, rather than simple theoretical priors that yield vacuous bounds.
Experimental setup
Metrics: generalization bound tightness
Builds on: PAC-Bayes
Closest papers by meaning
- Deterministic PAC-Bayesian generalization bounds for deep networks via generalizing noise-resilience (ICLR 2019 · similarity 0.90)
- On the Importance of Gradient Norm in PAC-Bayesian Bounds (NeurIPS 2022 · similarity 0.88)
- Dissecting Non-Vacuous Generalization Bounds based on the Mean-Field Approximation (ICML 2020 · similarity 0.88)
- Generalization Bounds via Meta-Learned Model Representations: PAC-Bayes and Sample Compression Hypernetworks (ICML 2025 · similarity 0.88)
- Recursive PAC-Bayes: A Frequentist Approach to Sequential Prior Updates with No Information Loss (NeurIPS 2024 · similarity 0.87)
- Compositional PAC-Bayes: Generalization of GNNs with persistence and beyond (NeurIPS 2024 · similarity 0.86)
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.