Research map / Optimization / Nonconvex and smooth optimization
Walking in the Shadow: A New Perspective on Descent Directions for Constrained Minimization
Hassan Mortagy, Swati Gupta, Sebastian Pokutta
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 |
|---|---|---|---|
| Use SHADOW-CG (FW plus shadow steps) instead of CG variants (AFW, PFW, DICG), assuming shadow oracle | SHADOW-CG · Video co-localization (YouTube-Objects), flow polytope | iteration count and wall-clock time | Improved |
| Use SHADOW-CG instead of CG variants without oracle access to the shadow | SHADOW-CG · Video co-localization (YouTube-Objects), flow polytope | iteration count | Improved |
| Use SHADOW-CG instead of PGD without oracle access | SHADOW-CG · Video co-localization (YouTube-Objects), flow polytope | wall-clock time | Improved |
| Use SHADOW-CG instead of PGD in iteration count | SHADOW-CG · Video co-localization (YouTube-Objects), flow polytope | iteration count | Hurt |
| Add FW steps to SHADOW-WALK (giving SHADOW-CG) to reduce iterations spent in TRACE (ablation) | SHADOW-CG vs SHADOW-WALK · Video co-localization (YouTube-Objects), flow polytope | iterations spent in TRACE: 10 -> 4 | Improved |
| Use approximate TRACE excluding in-face trace steps | SHADOW-CG · Video co-localization (YouTube-Objects), flow polytope | convergence | No clear effect |
| TRACE steps bounded by number of faces | SHADOW-WALK · Video co-localization (YouTube-Objects), flow polytope | TRACE iterations vs theory bound | Improved |
| Theory: SHADOW-WALK iteration complexity versus SHADOW-CG | SHADOW-WALK vs SHADOW-CG · Theory | iteration bound | Mixed |
Closest papers by meaning
- Directional Smoothness and Gradient Methods: Convergence and Adaptivity (NeurIPS 2024 · similarity 0.88)
- Toward a Unified Theory of Gradient Descent under Generalized Smoothness (ICML 2025 · similarity 0.87)
- No Wrong Turns: The Simple Geometry Of Neural Networks Optimization Paths (ICML 2024 · similarity 0.86)
- New Perspectives on the Polyak Stepsize: Surrogate Functions and Negative Results (NeurIPS 2025 · similarity 0.86)
- Exploring Landscapes for Better Minima along Valleys (NeurIPS 2025 · similarity 0.86)
- Block Acceleration Without Momentum: On Optimal Stepsizes of Block Gradient Descent for Least-Squares (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.
