Research map / Large language models / Agents, code and math
The Geometry of Reasoning: Flowing Logics in Representation Space
Yufa Zhou, Yixiao Wang, Xunjian Yin, Shuyan Zhou, Anru Zhang
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 |
|---|---|---|---|
| Compare first- and second-order geometry (velocity, curvature) instead of raw positions when grouping flows by logic | Reasoning-flow similarity · Logic dataset, Qwen3 0.6B | Similarity: position 0.26 vs velocity 0.17 / curvature 0.53 for logic; topic 0.07 / 0.11 | Improved |
| Randomly shuffle the order of logical steps (ablation) | Qwen3 0.6B reasoning flows · Logic dataset | Velocity and curvature similarity: velocity 0.17 -> 0.02; curvature 0.53 -> 0.02 | Hurt |
| Randomly shuffle the order of logical steps (ablation) | Qwen3 0.6B reasoning flows · Logic dataset | Position similarity: language 0.85 -> 0.81 | No clear effect |
| Scale model size (Qwen3 0.6B to 1.7B to 4B) (ablation) | Reasoning-flow similarity · Logic dataset | Velocity and curvature similarity: velocity 0.17, 0.19, 0.16 | No clear effect |
| Switch model family (Qwen1.5, Qwen2, Qwen3, LLaMA3) (ablation) | Reasoning-flow similarity · Logic dataset | Curvature similarity: 0.52, 0.52, 0.53, 0.58 | No clear effect |
Closest papers by meaning
- Geometric Latent Reasoning Induces Shorter Generations in LLMs (NeurIPS 2026 · similarity 0.89)
- AlgoTrace: Algorithmic Primitives and Compositional Geometry of Reasoning in Language Models (ICML 2026 · similarity 0.88)
- DAG-Math: Graph-Guided Mathematical Reasoning in LLMs (ICLR 2026 · similarity 0.88)
- Improving Reasoning Performance in Large Language Models via Representation Engineering (ICLR 2025 · similarity 0.88)
- Boule or Baguette? A Study on Task Topology, Length Generalization, and the Benefit of Reasoning Trace (NeurIPS 2026 · similarity 0.88)
- Search for Truth from Reasoning: A Dynamic Representation Editing Framework for Steering LLM Trajectories (ICML 2026 · similarity 0.87)
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.
