Research map / Optimization / Nonconvex and smooth optimization
Heavy Ball Momentum for Conditional Gradient
Bingcong Li, Alireza Sadeghi, Georgios B. Giannakis
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 heavy-ball momentum with δ_k = η_k = 2/(k+2) | Frank-Wolfe · Convex smooth optimization | primal-dual error bound: 4LD^2/(k+1) -> 2LD^2/(k+1) | Improved |
| Add restart when the standard FW gap is smaller than the generalized gap | HFW · Convex smooth optimization | primal-dual error bound: 2LD^2/(k+1) -> 2LD^2/(k + C^s) | Improved |
| Use uniform averaging of past gradients, δ_k = 1/(k+1), instead of the 2/(k+2) schedule (ablation) | HFW · Convex smooth optimization | primal-dual error bound: 2LD^2/(k+1) -> LD^2 ln (k+1)/(2k) | Hurt |
| Use a directional local Lipschitz constant instead of the global constant (ablation) | HFW smooth step size · Binary classification datasets | convergence speed: by an order of magnitude | Improved |
| Use WFW rather than UFW (ablation) | HFW · Binary classification, ℓ2-, ℓ1-, and n-support norm balls | primal error convergence | Improved |
| Use UFW rather than FW (ablation) | HFW · Binary classification, ℓ2-norm ball | primal error convergence | Mixed |
| Use UFW rather than FW (ablation) | HFW · Binary classification, n-support norm ball | primal error convergence | No clear effect |
| Use UFW-s rather than WFW-s (ablation) | HFW with smooth step sizes · Binary classification, n-support norm ball | primal error convergence | Hurt |
| Use directional smooth step sizes with UFW instead of WFW (ablation) | HFW · Binary classification datasets | primal error convergence | Hurt |
| Add restart (ablation) | WFW · Binary classification | primal error convergence | Improved |
| Use WFW rather than FW (ablation) | Frank-Wolfe · MovieLens100K matrix completion | optimality error | Improved |
| Use UFW-s rather than WFW-s (ablation) | HFW with smooth step sizes · MovieLens100K matrix completion | solution rank | Improved |
Closest papers by meaning
- Accelerated Convergence of Stochastic Heavy Ball Method under Anisotropic Gradient Noise (ICLR 2024 · similarity 0.88)
- Accelerated Over-Relaxation Heavy-Ball Method: Achieving Global Accelerated Convergence with Broad Generalization (ICLR 2025 · similarity 0.87)
- Heavy-Ball Momentum Method in Continuous Time and Discretization Error Analysis (NeurIPS 2025 · similarity 0.85)
- Efficient Quadratic Corrections for Frank-Wolfe Algorithms (NeurIPS 2025 · similarity 0.85)
- On the Performance Analysis of Momentum Method: A Frequency Domain Perspective (ICLR 2025 · similarity 0.85)
- Sarah Frank-Wolfe: Methods for Constrained Optimization with Best Rates and Practical Features (ICML 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.
