Research map / Learning theory and algorithms / Kernels and regression theory
More PAC-Bayes bounds: From bounded losses, to losses with general tail behaviors, to anytime validity
Borja Rodríguez-Gálvez, Ragnar Thobaben, Mikael Skoglund
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 the proposed fast/mixed-rate bound objective instead of the classical bound objective | PAC-Bayes with backprop · MNIST, CNN, PAC-Bayes with backprop | population risk certificate: 0.23658 -> 0.17501 | Improved |
| use the proposed bound objective instead of previous bound objective 1 | PAC-Bayes with backprop · MNIST, CNN, PAC-Bayes with backprop | population risk certificate: 0.20870 -> 0.17501 | Improved |
| use the proposed bound objective instead of previous bound objective 2 | PAC-Bayes with backprop · MNIST, CNN, PAC-Bayes with backprop | population risk certificate: 0.21159 -> 0.17501 | Improved |
| use the second proposed objective instead of previous bound objective 1 | PAC-Bayes with backprop · MNIST, CNN, PAC-Bayes with backprop | population risk certificate: 0.20870 -> 0.19763 | Improved |
| use the proposed objective instead of the classical objective | PAC-Bayes with backprop · MNIST, CNN, PAC-Bayes with backprop | empirical risk: 0.23658 -> 0.07054 | Improved |
| use the proposed objective instead of the classical objective | PAC-Bayes with backprop · MNIST, CNN, PAC-Bayes with backprop | normalized dependency D/n: 0.02715 -> 0.04649 | Hurt |
| use hard-coded Seeger-Langford gradients instead of the proposed fast-rate objective | PAC-Bayes with backprop · MNIST, CNN, PAC-Bayes with backprop | population risk certificate: 0.17501 -> 0.16922 | Improved |
Closest papers by meaning
- PAC-Bayes-Chernoff bounds for unbounded losses (NeurIPS 2024 · similarity 0.94)
- Block-Sample MAC-Bayes Generalization Bounds (ICLR 2026 · similarity 0.91)
- PAC-Bayes under potentially heavy tails (NeurIPS 2019 · similarity 0.89)
- Toward Better PAC-Bayes Bounds for Uniformly Stable Algorithms (NeurIPS 2023 · similarity 0.89)
- A Unified Recipe for Deriving (Time-Uniform) PAC-Bayes Bounds (NeurIPS 2024 · similarity 0.88)
- Controlling Multiple Errors Simultaneously with a PAC-Bayes Bound (NeurIPS 2024 · 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.
