Research map / Large language models / Agents, code and math
MCCE: A Framework for Multi-LLM Collaborative Search in Discrete Spaces with Similarity-Filtered Preference Learning
Nian Ran, Zhongzheng Li, Yue Wang, Qingsong Ran, Xiaoyuan Zhang, Shikun Feng, Richard Allmendinger, Xiaoguang Zhao
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 |
|---|---|---|---|
| replace SFT with DPO for refining the local model (ablation) | MCCE · 5-objective molecular design | hypervolume: 0.709 -> 0.847 | Improved |
| replace RL with DPO for refining the local model (ablation) | MCCE · 5-objective molecular design | hypervolume: 0.709 -> 0.847 | Improved |
| add parameter training of the local model (DPO) to plain two-LLM collaboration (ablation) | MCCE · 5-objective molecular design | hypervolume: 0.695 -> 0.847 | Improved |
| add the frozen API LLM to a lone local Qwen2.5-7B-Instruct (ablation) | MCCE · 5-objective molecular design | hypervolume: 0.516 -> 0.847 | Improved |
| replace trainable local model with a second frozen API (GPT-4o + Gemini) (ablation) | MCCE · 5-objective molecular design | hypervolume: 0.847 -> 0.492 | Hurt |
| use only the relaxed similarity window I3 instead of nested windows (ablation) | MCCE similarity filtering · 5-objective molecular design | Top1 fitness: 4.35 -> 4.32 | No clear effect |
| use embedding-based similarity instead of fingerprint similarity (ablation) | MCCE similarity filtering · 5-objective molecular design | Top1 fitness: 4.35 -> 4.29 | Hurt |
| vary score quantile alpha from 0.30 to 0.20 or 0.40 (ablation) | MCCE · 5-objective molecular design | hypervolume: 0.847 -> 0.828 (0.20), 0.843 (0.40) | No clear effect |
| use 50/16 API/local call split instead of 50/32 (ablation) | MCCE · 5-objective molecular design | hypervolume: 0.847 -> 0.734 | Hurt |
| use 50/50 API/local call split instead of 50/32 (ablation) | MCCE · 5-objective molecular design | Top1 fitness: 4.35 -> 4.31 | Hurt |
| update the local model less often (every 500 candidates instead of one round) (ablation) | MCCE · 5-objective molecular design | hypervolume: 0.847 -> 0.796 | Hurt |
| update every 200 candidates instead of default (ablation) | MCCE · 5-objective molecular design | hypervolume: 0.847 -> 0.856 | No clear effect |
Closest papers by meaning
- EVOLvE: Evaluating and Optimizing LLMs For In-Context Exploration (ICML 2025 · similarity 0.87)
- Evolution Fine-Tuning: Learning to Discover Across 371 Optimization Tasks (NeurIPS 2026 · similarity 0.87)
- Efficient Evolutionary Search Over Chemical Space with Large Language Models (ICLR 2025 · similarity 0.87)
- Model Swarms: Collaborative Search to Adapt LLM Experts via Swarm Intelligence (ICML 2025 · similarity 0.87)
- HSEvo: Elevating Automatic Heuristic Design with Diversity-Driven Harmony Search and Genetic Algorithm Using LLMs (AAAI 2025 · similarity 0.86)
- ReEvo: Large Language Models as Hyper-Heuristics with Reflective Evolution (NeurIPS 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.
