Research map / Large language models / Agents, code and math
MindFlow: Mind Supernet Powered Thinking Flows for Research Idea Innovation
Mengdi Liu, Wenjue Chen, Wenyue Chen, Cheng Yang, Fanqi Kong, Zhangyang Gao, Xiaoxue Cheng, Yiheng Li, Yujian Yuan, Keliang Li, Hong Chang, Shiguang Shan and 1 more
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 supernet controller instead of fixed Critical Thinking (ablation) | MindFlow · IdeaBench (LLM-judged win rate) | overall win rate: 0.336 -> 0.541 | Improved |
| add supernet controller instead of fixed Constraint-Driven Thinking (ablation) | MindFlow · IdeaBench (LLM-judged win rate) | overall win rate: 0.472 -> 0.541 | Improved |
| add supernet controller instead of fixed Counterfactual Thinking (ablation) | MindFlow · IdeaBench (LLM-judged win rate) | overall win rate: 0.426 -> 0.541 | Improved |
| add supernet controller instead of fixed Analogical Thinking (ablation) | MindFlow · IdeaBench (LLM-judged win rate) | overall win rate: 0.438 -> 0.541 | Improved |
| add supernet controller instead of fixed Divergent + Convergent Thinking (ablation) | MindFlow · IdeaBench (LLM-judged win rate) | overall win rate: 0.383 -> 0.541 | Improved |
| add supernet controller instead of fixed Counterfactual Thinking on problem finding novelty (ablation) | MindFlow · IdeaBench (LLM-judged win rate) | problem finding novelty: 0.901 -> 0.744 | Hurt |
| add supernet controller instead of fixed Critical Thinking on problem solving feasibility (ablation) | MindFlow · IdeaBench (LLM-judged win rate) | problem solving feasibility: 0.275 -> 0.257 | Mixed |
| use MindFlow instead of VIRSCI on problem solving feasibility | MindFlow · IdeaBench (LLM-judged win rate) | problem solving feasibility: 0.109 -> 0.257 | Improved |
| use MindFlow instead of VIRSCI overall (LLM judged) | MindFlow · IdeaBench (LLM-judged win rate) | overall win rate: 0.470 -> 0.541 | Improved |
| use MindFlow instead of VIRSCI overall (human experts) | MindFlow · IdeaBench (human expert, 50 queries) | overall win rate: 0.475 -> 0.580 | Improved |
| use MindFlow instead of VIRSCI on Science domain | MindFlow · IdeaBench Science domain | overall win rate: 0.473 -> 0.412 | Hurt |
| use MindFlow instead of VIRSCI on Robotics domain | MindFlow · IdeaBench Robotics domain | overall win rate: 0.533 -> 0.694 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- AI-Researcher — AI-Researcher: Autonomous Scientific Innovation (NeurIPS 2025) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- MindGYM: What Matters in Question Synthesis for Thinking-Centric Fine-Tuning? (NeurIPS 2025 · similarity 0.85)
- MindLoom: Composing Thought Modes for Frontier-Level Reasoning Data Synthesis (NeurIPS 2026 · similarity 0.84)
- IDEAFix: Evaluation Framework for Creative Defixation Prompting in LLMs (NeurIPS 2026 · similarity 0.84)
- InnovatorBench: Evaluating Agents’ Ability to Conduct Innovative AI Research (ICLR 2026 · similarity 0.83)
- DyFlow: Dynamic Workflow Framework for Agentic Reasoning (NeurIPS 2025 · similarity 0.83)
- AI-Researcher: Autonomous Scientific Innovation (NeurIPS 2025 · similarity 0.83)
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.
