Research map / Optimization / Nonconvex and smooth optimization
A Stochastic Bundle Method for Interpolation
Alasdair Paren, Leonard Berrada, Rudra P. K. Poudel, M. Pawan Kumar
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 |
|---|---|---|---|
| replace SGD constant step size with BORAT3 (ablation) | WRN · Tiny ImageNet val (CE loss) | validation accuracy: 59.4 -> 61.1 | Improved |
| use BORAT5 instead of ALI-G (ablation) | WRN · Tiny ImageNet val (CE loss) | validation accuracy: 61.1 -> 62.1 | Improved |
| use BORAT3 instead of ALI-G with hinge loss (ablation) | WRN · Tiny ImageNet val (hinge loss) | validation accuracy: 24.9 -> 44.1 | Improved |
| add bundle size 3 to ALI-G (ablation) | DenseNet · CIFAR-100 | test accuracy: 76.2 -> 76.5 | Improved |
| add bundle size 3 to ALI-G (ablation) | WRN · CIFAR-100 | test accuracy: 76.1 -> 76.0 | No clear effect |
| use BORAT5 instead of BORAT3 (ablation) | WRN · CIFAR-10 | test accuracy: 95.4 -> 95.0 | Hurt |
| use BORAT3 instead of constant-step SGD (ablation) | WRN · CIFAR-100 | test accuracy: 67.8 -> 76.0 | Improved |
| use ALI-G instead of Adam (ablation) | Bi-LSTM · SNLI (CE loss) | test accuracy: 84.5 -> 84.8 | Improved |
| use BORAT instead of Adam with hinge loss (ablation) | WRN · Tiny ImageNet val | validation accuracy: 2.4 -> 44.1 | Improved |
| use BORAT3 instead of tuned-schedule SGD (ablation) | WRN · SVHN | test accuracy: 98.3 -> 98.1 | No clear effect |
| increase bundle size N beyond 1 (more linear approximations) (ablation) | BORAT · hyperparameter grid, CIFAR100 and Tiny ImageNet | robustness to eta and r | Improved |
Closest papers by meaning
- Training Neural Networks for and by Interpolation (ICML 2020 · similarity 0.87)
- Neumann Optimizer: A Practical Optimization Algorithm for Deep Neural Networks (ICLR 2018 · similarity 0.86)
- No Wrong Turns: The Simple Geometry Of Neural Networks Optimization Paths (ICML 2024 · similarity 0.85)
- Simplicity Bias and Optimization Threshold in Two-Layer ReLU Networks (ICML 2025 · similarity 0.85)
- Deconstructing the Goldilocks Zone of Neural Network Initialization (ICML 2024 · similarity 0.84)
- A Stochastic Approach to Bi-Level Optimization for Hyperparameter Optimization and Meta Learning (AAAI 2025 · similarity 0.84)
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.
