Research map / Learning theory and algorithms / Prediction and decision losses
Minding the Gaps for Block Frank-Wolfe Optimization of Structured SVMs
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| adaptive gap-based sampling of objects (probability proportional to stored block gap) instead of uniform sampling (ablation) | BCFW · OCR, HorseSeg, LSP, CoNLL experiments | duality gap vs effective passes and time | Helped |
| caching oracle calls with gap-based cache-hit criterion (ablation) | BCFW and BCPFW · OCR, HorseSeg, LSP, CoNLL experiments | duality gap vs effective passes | Helped |
| caching oracle calls with gap-based cache-hit criterion (ablation) | BCFW and BCPFW · expensive max oracle experiments | duality gap vs time | Helped |
| pairwise steps (BCPFW) instead of BCFW steps (ablation) | BCFW · OCR, HorseSeg, LSP, CoNLL experiments | duality gap | Helped |
| epsilon-approximate regularization path (eps=0.1) vs grid search of 31 lambda values with warm start | SSVM regularization path · HorseSeg-small | cumulative running time to cover lambda values | Mixed |
| gap sampling with exact block gaps vs uniform sampling (toy example) | BCFW · synthetic toy SSVM example | number of oracle calls | Helped |
Closest papers by meaning
- A Smoother Way to Train Structured Prediction Models (NeurIPS 2018 · similarity 0.89)
- Parallel and Distributed Block-Coordinate Frank-Wolfe Algorithms (ICML 2016 · similarity 0.87)
- Primal-Dual Block Generalized Frank-Wolfe (NeurIPS 2019 · similarity 0.86)
- Faster Algorithms for Structured Linear and Kernel Support Vector Machines (ICLR 2025 · similarity 0.85)
- Sarah Frank-Wolfe: Methods for Constrained Optimization with Best Rates and Practical Features (ICML 2024 · similarity 0.83)
- Structured IB: Improving Information Bottleneck with Structured Feature Learning (AAAI 2025 · similarity 0.83)
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.