Research map / Reinforcement learning / Policy gradient and actor-critic
Learning to Perform Local Rewriting for Combinatorial Optimization
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 hand-tuned local heuristics with learned region and rewrite-rule selection | NeuRewriter on Halide expressions · Halide expression simplification test | average expression length reduction: 50.81 -> 57.28 | Helped |
| Replace hand-tuned local heuristics with learned region and rewrite-rule selection | NeuRewriter on Halide expressions · Halide expression simplification test | average tree size reduction: 15.82 -> 16.71 | Helped |
| Learn when to apply uphill rewriting rules | NeuRewriter on Halide expressions · Halide expression simplification test | average expression length reduction | Helped |
| Iteratively rewrite a complete schedule instead of constructing the schedule from scratch | NeuRewriter on job scheduling · Job scheduling, D=20 | average job slowdown: 10.18 -> 3.94 | Helped |
| Use a locally rewritten schedule rather than offline SJF planning | NeuRewriter on job scheduling · Job scheduling, D=20 | average job slowdown: 4.26 -> 3.94 | Helped |
| Train on dynamic rather than steady job frequency (ablation) | NeuRewriter · Job scheduling, dynamic job frequency | average job slowdown: 2.59 -> 2.56 | Helped |
| Train on dynamic rather than steady job frequency (ablation) | NeuRewriter · Job scheduling, steady job frequency | average job slowdown: 3.94 -> 3.99 | Hurt |
| Train on uniform rather than non-uniform job resources (ablation) | NeuRewriter · Job scheduling, uniform job resources | average job slowdown: 3.13 -> 2.73 | Helped |
| Train on uniform rather than non-uniform job resources (ablation) | NeuRewriter · Job scheduling, non-uniform job resources | average job slowdown: 3.94 -> 4.05 | Hurt |
| Iteratively rewrite an initial route rather than construct a route from scratch | NeuRewriter on CVRP · CVRP, VRP20 Cap30 | average tour length: 6.25 -> 6.16 | Helped |
| Use local rewriting rather than OR-Tools route optimization | NeuRewriter on CVRP · CVRP, VRP100 Cap50 | average tour length: 17.16 -> 16.10 | Helped |
Closest papers by meaning
- Preference Optimization for Combinatorial Optimization Problems (ICML 2025 · similarity 0.88)
- Improving Generalization of Neural Combinatorial Optimization for Vehicle Routing Problems via Test-Time Projection Learning (NeurIPS 2025 · similarity 0.87)
- GOAL: A Generalist Combinatorial Optimization Agent Learner (ICLR 2025 · similarity 0.86)
- Neural Solver Selection for Combinatorial Optimization (ICML 2025 · similarity 0.86)
- Large Language Models as End-to-end Combinatorial Optimization Solvers (NeurIPS 2025 · similarity 0.86)
- Teacher-Aware Evolution of Heuristic Programs from Learned Optimization Policies (NeurIPS 2026 · 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.