Research map / Large language models / Agents, code and math
CALM: Co-evolution of Algorithms and Language Model for Automatic Heuristic Design
Ziyao Huang, Weiwei Wu, Kui Wu, Wei-Bin Lee, Jianping Wang
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 GRPO fine-tuning of the local LLM (Qwen INT4 vs GPT-4o-mini API without GRPO) | CALM · CVRP N=50 (ACO) | optimality gap (%): 5.81% -> 3.83% | Improved |
| add GRPO fine-tuning of the local LLM | CALM · OP N=200 (ACO) | optimality gap (%): 16.22% -> 12.58% | Improved |
| add GRPO fine-tuning of the local LLM | CALM · OBP average | optimality gap (%): 0.82% -> 0.71% | Improved |
| add GRPO fine-tuning of the local LLM | CALM · TSP N=200 | optimality gap (%): 13.56% -> 13.41% | Improved |
| remove GRPO module (disable RL) (ablation) | CALM · OBP, CVRP, OP, TSP ablations | optimality gap (%) | Improved |
| use CALM verbal guidance alone with GPT-4o-mini API versus MCTS-AHD | CALM (API, no GRPO) · OBP average | optimality gap (%): 0.89% -> 0.82% | Improved |
| use CALM's reward instead of performance-based reward (ablation) | CALM · OP | optimality gap (%) | Improved |
| use CALM's reward instead of the {0.5 r_rand, 1} improvement reward (ablation) | CALM · OBP and OP | optimality gap (%) | Improved |
| add the collapse mechanism (ablation) | CALM · OBP and OP | optimality gap (%) | Improved |
| add collapse with strictest tolerance (delta_0=0.005, C=15) (ablation) | CALM · OP | optimality gap (%) | Hurt |
| add the simplification operator (ablation) | CALM · OBP and OP | optimality gap (%) | Improved |
| add diversity-aware selection in crossover (ablation) | CALM · OBP and OP | optimality gap (%) | Improved |
Closest papers by meaning
- ReEvo: Large Language Models as Hyper-Heuristics with Reflective Evolution (NeurIPS 2024 · similarity 0.91)
- HSEvo: Elevating Automatic Heuristic Design with Diversity-Driven Harmony Search and Genetic Algorithm Using LLMs (AAAI 2025 · similarity 0.89)
- EvoOptiGraph: Weakness-Driven Coevolution via Graph-Based Structural Generation for Optimization Modeling (NeurIPS 2026 · similarity 0.89)
- Evolution Fine-Tuning: Learning to Discover Across 371 Optimization Tasks (NeurIPS 2026 · similarity 0.89)
- Prompted Policy Search: Reinforcement Learning through Linguistic and Numerical Reasoning in LLMs (NeurIPS 2025 · similarity 0.89)
- PathWise: Planning through World Model for Automated Heuristic Design via Self-Evolving LLMs (ICML 2026 · similarity 0.89)
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.
