Research map / Large language models / Retrieval and knowledge
FutureMind: Equipping Small Language Models with Strategic Thinking-Pattern Priors via Adaptive Knowledge Distillation
Shaoxiong Yang, Junting Li, Mengyuan Zhang, Chao Li, Wei Liu, Jian Luan
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 FutureMind planning to ToolCall agent | Qwen2.5-3B student · Multi-hop QA avg (4 benchmarks) | ACC_E: 24.39 -> 32.16 (vs Search-o1) | Improved |
| add FutureMind planning | Llama3.1-8B · Bamboogle | ACC_L: 52.00 -> 64.00 (vs Search-o1) | Improved |
| use 3B model as teacher for 3B student (ablation) | TC+FM · Multi-hop QA avg (4 benchmarks) (student Qwen2.5-3B) | ACC_L: 29.49 -> 20.03 | Hurt |
| use 14B teacher instead of 72B teacher (ablation) | TC+FM, Qwen2.5-7B student · Multi-hop QA avg (4 benchmarks) | ACC_L: 44.38 -> 46.83 | Improved |
| add Strategy Planning module (ablation) | FutureMind Qwen2.5-7B · Multi-hop QA avg (4 benchmarks) | ACC_L: 40.33 -> 46.83 | Improved |
| add Problem Analysis module (ablation) | FutureMind Qwen2.5-7B · Multi-hop QA avg (4 benchmarks) | ACC_L: 43.84 -> 46.83 | Improved |
| add Logical Reasoning module (ablation) | FutureMind Qwen2.5-7B · Multi-hop QA avg (4 benchmarks) | ACC_L: 42.45 -> 46.83 | Improved |
| add Retrieval Guidance module (ablation) | FutureMind Qwen2.5-7B · Multi-hop QA avg (4 benchmarks) | ACC_L: 43.59 -> 46.83 | Improved |
| add Forward Stepwise strategy (ablation) | FutureMind Qwen2.5-7B · Multi-hop QA avg (4 benchmarks) | ACC_L: 42.79 -> 46.83 | Improved |
| add Backward Constraint strategy (ablation) | FutureMind Qwen2.5-7B · Multi-hop QA avg (4 benchmarks) | ACC_L: 44.67 -> 46.83 | Improved |
| add Parallel Intersection strategy (ablation) | FutureMind Qwen2.5-7B · Multi-hop QA avg (4 benchmarks) | ACC_L: 43.64 -> 46.83 | Improved |
| use Qwen2.5-72B teacher instead of Llama3.1-70B (ablation) | TC+FM, Qwen2.5-72B student · Multi-hop QA avg (4 benchmarks) | ACC_L: 55.68 -> 63.25 | Improved |
Closest papers by meaning
- RATT: A Thought Structure for Coherent and Correct LLM Reasoning (AAAI 2025 · similarity 0.90)
- Think or Not? Exploring Thinking Efficiency in Large Reasoning Models via an Information-Theoretic Lens (NeurIPS 2025 · similarity 0.89)
- Algorithm of Thoughts: Enhancing Exploration of Ideas in Large Language Models (ICML 2024 · similarity 0.89)
- Simple is Effective: The Roles of Graphs and Large Language Models in Knowledge-Graph-Based Retrieval-Augmented Generation (ICLR 2025 · similarity 0.89)
- Don’t Think Longer, Think Wisely: Optimizing Thinking Dynamics for Large Reasoning Models (NeurIPS 2025 · similarity 0.89)
- Reasoning Scaffolding: Distilling the Flow of Thought from LLMs (ICLR 2026 · similarity 0.88)
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.
