Research map / Optimization / Optimizers
Polygonal Unadjusted Langevin Algorithms: Creating stable and efficient adaptive algorithms for neural networks
Dong-Young Lim, Sotirios Sabanis
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 |
|---|---|---|---|
| add boosting function (finite epsilon) (ablation) | TheoPouLa · CIFAR-10 VGG11 | test accuracy: 91.48 -> 92.30 | Improved |
| add boosting function (finite epsilon) (ablation) | TheoPouLa · CIFAR-10 ResNet34 | test accuracy: 94.31 -> 95.43 | Improved |
| add boosting function (finite epsilon) (ablation) | TheoPouLa · CIFAR-10 DenseNet121 | test accuracy: 94.22 -> 95.66 | Improved |
| add boosting function (finite epsilon) (ablation) | TheoPouLa · CIFAR-100 VGG11 | test accuracy: 68.11 -> 70.31 | Improved |
| add boosting function (finite epsilon) (ablation) | TheoPouLa · CIFAR-100 ResNet34 | test accuracy: 75.91 -> 77.60 | Improved |
| add boosting function (finite epsilon) (ablation) | TheoPouLa · CIFAR-100 DenseNet121 | test accuracy: 77.99 -> 79.90 | Improved |
| lambda 0.5 instead of 0.1 (too large step) (ablation) | TheoPouLa · CIFAR-100 ResNet34 | test accuracy: 74.02 -> 1.90 | Hurt |
| lambda 0.01 instead of 0.1 (ablation) | TheoPouLa · CIFAR-10 VGG11 | test accuracy: 92.14 -> 89.89 | Hurt |
| epsilon 0.001 instead of 0.1 (ablation) | TheoPouLa · CIFAR-10 VGG11 | test accuracy: 92.14 -> 26.03 | Hurt |
| epsilon 1 instead of 0.1 (ablation) | TheoPouLa · CIFAR-100 VGG11 | test accuracy: 70.57 -> 68.75 | Hurt |
| inverse temperature beta 10^4 instead of 10^10 (ablation) | TheoPouLa · CIFAR-100 VGG11 | test accuracy: 70.16 -> 20.69 | Hurt |
| inverse temperature beta 10^8 vs 10^12 (ablation) | TheoPouLa · CIFAR-10 ResNet34 | test accuracy: 95.42 -> 95.43 | No clear effect |
Closest papers by meaning
- A General System of Differential Equations to Model First-Order Adaptive Algorithms (JMLR 2020 · similarity 0.87)
- Global Convergence of Langevin Dynamics Based Algorithms for Nonconvex Optimization (NeurIPS 2018 · similarity 0.86)
- When is the Convergence Time of Langevin Algorithms Dimension Independent? A Composite Optimization Viewpoint (NeurIPS 2022 · similarity 0.86)
- Stochastic Gradient and Langevin Processes (ICML 2020 · similarity 0.86)
- Improved Convergence Rate of Stochastic Gradient Langevin Dynamics with Variance Reduction and its Application to Optimization (NeurIPS 2022 · similarity 0.86)
- Learning Multi-Index Models with Neural Networks via Mean-Field Langevin Dynamics (ICLR 2025 · 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.
