Research map / Optimization / Nonconvex and smooth optimization
Lazifying Conditional Gradient Algorithms
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 |
|---|---|---|---|
| Replace linear optimization calls at every iteration with weak separation using cached solutions and early termination | Frank-Wolfe and conditional gradient algorithms · Offline and online optimization experiments | wall-clock time | Helped |
| Use lazy Frank-Wolfe instead of standard Frank-Wolfe | Vanilla Frank-Wolfe · Matrix completion | wall-clock time to reach lower function values | Helped |
| Use lazy Frank-Wolfe instead of standard Frank-Wolfe | Vanilla Frank-Wolfe · Large netgen path-polytope instances | dual bound / Wolfe gap | Helped |
| Use lazy pairwise conditional gradient instead of pairwise conditional gradient | Pairwise Conditional Gradient · MIPLIB structured regression instances | wall-clock time | Helped |
| Use lazy online conditional gradient instead of online Frank-Wolfe | Online Conditional Gradient · TSP polytope, 11-node graph, quadratic losses | final loss | Helped |
| Use lazy online conditional gradient instead of online Frank-Wolfe | Online Conditional Gradient · Cut polytope, 23-node graph, quadratic losses | wall-clock time | Helped |
| Use weak separation rather than linear optimization at every iteration | Lazy conditional gradient algorithms · Iterations across the reported experiments | convergence by iteration count | No clear effect |
| Use lazy online conditional gradient rather than online Frank-Wolfe | Online Conditional Gradient · TSP polytope, 16-node graph, quadratic losses | iterations completed within the time limit | Mixed |
| Use lazy Frank-Wolfe instead of standard Frank-Wolfe | Vanilla Frank-Wolfe · Matrix completion, first reported pair of instances | iterations | Hurt |
| Use caching in the weak separation oracle | Lazy conditional gradient algorithms · Most computational experiments | cache hit rate | Helped |
| Use the parameter-free lazy variant rather than a predetermined progress schedule | Lazy Frank-Wolfe · Offline optimization experiments | convergence speed | Helped |
| Use early termination even when caching has low hit rate | Lazy Online Conditional Gradient · TSP polytope, 16-node graph, quadratic losses | wall-clock time | Helped |
Closest papers by meaning
- Frank-Wolfe with Subsampling Oracle (ICML 2018 · similarity 0.91)
- Blended Conditonal Gradients (ICML 2019 · similarity 0.88)
- Beyond Short Steps in Frank-Wolfe Algorithms (ICLR 2026 · similarity 0.88)
- Stochastic Frank-Wolfe for Composite Convex Minimization (NeurIPS 2019 · similarity 0.88)
- Efficient Quadratic Corrections for Frank-Wolfe Algorithms (NeurIPS 2025 · similarity 0.87)
- Sarah Frank-Wolfe: Methods for Constrained Optimization with Best Rates and Practical Features (ICML 2024 · similarity 0.86)
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.