Research map / Learning theory and algorithms / Kernels and regression theory
Generalization of Gibbs and Langevin Monte Carlo Algorithms in the Interpolation Regime
Andreas Maurer, Erfan Mirzaei, Massimiliano Pontil
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 calibrated Gibbs-posterior bound instead of self-certified f_quad certificate | FCN on MNIST · MNIST 10-class FCN | true 0-1 risk bound: 0.3155 -> 0.2470 | Improved |
| use calibrated Gibbs-posterior bound instead of f_quad certificate | CNN on MNIST · MNIST 10-class CNN | true 0-1 risk bound: 0.2165 -> 0.2025 | Improved |
| use SGLD-trained standard model instead of f_bbb | FCN on MNIST · MNIST 10-class FCN | true 0-1 risk bound: 0.5516 -> 0.2470 | Improved |
| use SGLD-trained standard model instead of f_bbb | FCN on MNIST · MNIST 10-class FCN | test 0-1 error: 0.0293 -> 0.0338 | Hurt |
| use SGLD-trained standard model instead of f_bbb | CNN on MNIST · MNIST 10-class CNN | test 0-1 error: 0.0154 -> 0.0182 | Hurt |
| lower temperature (beta 500 to 1000) in SGLD with true labels | SGLD fully connected net · MNIST binary | train error: 0.17 -> 0.1 | Improved |
| use random labels instead of true labels at beta=64000 | SGLD · MNIST binary | train error and bound: train error about 0.1 while bound above 0.5 | Mixed |
Closest papers by meaning
- Temperature is All You Need for Generalization in Langevin Dynamics and other Markov Processes (NeurIPS 2025 · similarity 0.89)
- An Exact Characterization of the Generalization Error for the Gibbs Algorithm (NeurIPS 2021 · similarity 0.88)
- On Generalization Error Bounds of Noisy Gradient Methods for Non-Convex Learning (ICLR 2020 · similarity 0.88)
- Non-asymptotic Error Bounds in $\mathcal{W}_2$-Distance with Sqrt(d) Dimension Dependence and First Order Convergence for Langevin Monte Carlo beyond Log-Concavity (ICML 2025 · similarity 0.87)
- Langevin Monte Carlo for strongly log-concave distributions: Randomized midpoint revisited (ICLR 2024 · similarity 0.85)
- Constrained Sampling with Primal-Dual Langevin Monte Carlo (NeurIPS 2024 · similarity 0.85)
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.
