Research map / Large language models / Agents, code and math
AutoEP: LLMs-Driven Automation of Hyperparameter Evolution for Metaheuristic Algorithms
Zhenxing Xu, Yizhe Zhang, Weidong Bao, Hao Wang, Ming Chen, Haoran Ye, Wenzheng Jiang, Hui Yan, Ji 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 ELA features to the LLM input (ablation) | AutoEP on GA-2opt · TSPLIB dsj1000 | optimality gap (%): 6.46 -> 3.58 | Improved |
| use the multi-LLM Chain of Reasoning instead of a single LLM (ablation) | AutoEP on GA-2opt · TSPLIB dsj1000 | optimality gap (%): 7.11 -> 3.58 | Improved |
| include ELA and CoR together versus neither (ablation) | AutoEP on GA-2opt · TSPLIB dsj1000 | optimality gap (%): 7.93 -> 3.58 | Improved |
| replace Qwen3-30B CoR with a single large proprietary reasoning LLM (GPT-o1) (ablation) | AutoEP · TSPLIB eil51 | runtime (min): 5.8 -> 44.7 | Hurt |
| replace Qwen3-30B CoR with single GPT-o1 (ablation) | AutoEP · TSPLIB dsj1000 | optimality gap (%): 1.58 -> 1.59 | No clear effect |
| limit experience pool to sliding window of 20 iterations instead of full history (ablation) | AutoEP on GA-2opt · TSP-100 | optimality gap (%): 0.17 -> 0.01 | Improved |
| shrink experience window from 20 to 5 (ablation) | AutoEP on GA-2opt · TSP-100 | optimality gap (%): 0.01 -> 0.04 | Hurt |
| enlarge experience window from 20 to 50 (ablation) | AutoEP on GA-2opt · TSP-100 | optimality gap (%): 0.01 -> 0.03 | Hurt |
| adjust hyperparameters every 3-5 iterations instead of every iteration (ablation) | AutoEP · UAV-300 | convergence speed | Mixed |
| add AutoEP to GA-2opt | GA-2opt · TSPLIB dsj1000 | optimality gap (%): 7.14 -> 3.58 | Improved |
| add AutoEP on top of EoH or ReEvo enhanced GA-2opt | GA-2opt+EoH, GA-2opt+ReEvo · TSPLIB dsj1000 | optimality gap (%): 5.83 -> 3.61 (EoH) | Improved |
| use smaller open LLMs in AutoEP versus EoH and ReEvo | AutoEP · GA-2opt, various LLMs | optimality gap | Improved |
Closest papers by meaning
- HSEvo: Elevating Automatic Heuristic Design with Diversity-Driven Harmony Search and Genetic Algorithm Using LLMs (AAAI 2025 · similarity 0.88)
- ReEvo: Large Language Models as Hyper-Heuristics with Reflective Evolution (NeurIPS 2024 · similarity 0.87)
- Large Language Models to Enhance Bayesian Optimization (ICLR 2024 · similarity 0.86)
- LLMs Improving LLMs: Agentic Discovery for Test-Time Scaling (NeurIPS 2026 · similarity 0.86)
- AutoTool: Dynamic Tool Selection and Integration for Agentic Reasoning (ICML 2026 · similarity 0.86)
- REASONING COMPILER: LLM-Guided Optimizations for Efficient Model Serving (NeurIPS 2025 · 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.
