Research map / Optimization / Optimizers
Learning Proximal Operators to Discover Multiple Optima
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 |
|---|---|---|---|
| learned proximal operator with proximal term (vs. learned gradient descent operator GOL) | multi-solution object detection · COCO2017 | WD_inf and WP_inf^0.1: WD 0.251 -> 0.149; WP 0.243 -> 0.590 | Helped |
| learned proximal operator (vs. fixed-number chamfer prediction FN) | multi-solution object detection · COCO2017 | precision: 0.139 -> 0.817 | Helped |
| learned proximal operator vs. Faster R-CNN with 40.7% more parameters | multi-solution object detection · COCO2017 | WD_inf: 0.149 vs 0.140 (FRCNN .80) | Hurt |
| proximal term with lambda above weak-convexity constant (lambda=400 vs lambda=10 on 2D cosine) (ablation) | POL on 2D cosine function · 2D cosine (qualitative) | recovered optima | Helped |
| removing proximal term from learned operator | learned L2O operator for multi-solution optimization · toy example (argument) | solution diversity | Hurt |
| learned proximal operator (POL) vs. learned gradient operator (GOL) on conic sections | level set sampling · conic sections | WP at small delta | Helped |
| learned proximal operator (POL) vs. GOL on non-convex sparse recovery | sparse recovery with l_p norm · non-convex sparse recovery | objective quality | Helped |
| learned proximal operator (POL) vs. GOL on MCP sparse recovery | sparse recovery with MCP regularizer · MCP sparse recovery | objective histograms | Helped |
| learned proximal operator (POL) vs. proximal gradient descent (PGD) for p=1/2 | non-convex sparse recovery · p=1/2, alpha=0.627 | convergence speed | Helped |
| learned proximal operator (POL) vs. GOL on symmetry detection | 3D reflection symmetry detection · MCB symmetry detection | witnessed precision | Helped |
| learned proximal operator (POL) vs. GOL on max-cut rank-2 relaxation | max-cut on K_8 · max-cut | witnessed metrics | Helped |
| learned proximal operator (POL) vs. GOL on conic sections: GOL convergence speed at test time | level set sampling · conic sections | iterations to converge: POL <5 steps vs GOL slower (100 iterations used) | Helped |
Closest papers by meaning
- Inexact Proximal Gradient Methods for Non-Convex and Non-Smooth Optimization (AAAI 2018 · similarity 0.90)
- Operator Splitting with Hamilton-Jacobi-based Proximals (ICML 2026 · similarity 0.88)
- What's in a Prior? Learned Proximal Networks for Inverse Problems (ICLR 2024 · similarity 0.88)
- RECAPP: Crafting a More Efficient Catalyst for Convex Optimization (ICML 2022 · similarity 0.88)
- Adaptive Proximal Gradient Method for Convex Optimization (NeurIPS 2024 · similarity 0.88)
- Adaptive Proximal Gradient Methods Are Universal Without Approximation (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.