Research map / Large language models / Language, safety and interpretability
Beyond Scalars: Evaluating and Understanding LLM Reasoning via Geometric Progress and Stability
Xinyan Jiang, Ninghao Liu, Di Wang, Lijie Hu
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 curvature to displacement-only features (ablation) | TRACED · GPQA | AUROC: 0.7812 -> 0.8300 | Improved |
| add displacement to curvature-only features (ablation) | TRACED · GPQA | AUROC: 0.7244 -> 0.8300 | Improved |
| add both signatures vs displacement only (ablation) | TRACED · GSM8K | AUROC: 0.7634 -> 0.8061 | Improved |
| add reasoning-quality subspace projection (ablation) | TRACED · GPQA (DeepSeek-R1-Llama-8B) | AUPR: 0.624 -> 0.661 | Improved |
| add reasoning-quality subspace projection (ablation) | TRACED · GSM8K (DeepSeek-R1-Llama-8B) | AUPR: 0.813 -> 0.828 | Improved |
| add TRACED geometric score to GRPO reward | GRPO Qwen2.5-0.5B · GSM8K | accuracy (%): 39.65 -> 43.97 | Improved |
| add TRACED geometric score to GRPO reward | GRPO Qwen2.5-0.5B · MATH | accuracy (%): 28.33 -> 30.65 | Improved |
| add TRACED geometric score to GRPO reward | GRPO Qwen2.5-1.5B · GSM8K | accuracy (%): 67.83 -> 71.22 | Improved |
| add TRACED geometric score to GRPO reward | GRPO Qwen2.5-1.5B · MATH | accuracy (%): 47.20 -> 50.10 | Improved |
| apply centroid alignment before cross-domain transfer | TRACED · cross-domain transfer | AUPR | Improved |
| increase subspace dimension k up to 8 (ablation) | TRACED · four models | AUROC | Improved |
| vary reasoning difficulty (Easy to Hard) (ablation) | TRACED · DeepSeek-R1-Llama-8B, stratified | AUROC: 0.775 / 0.748 / 0.766 | No clear effect |
Closest papers by meaning
- How Far Ahead Do LLMs Plan? Uncovering the Latent Horizon in Chain-of-Thought Reasoning (ICML 2026 · similarity 0.88)
- The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity (NeurIPS 2025 · similarity 0.87)
- Demystifying Reasoning Dynamics with Mutual Information: Thinking Tokens are Information Peaks in LLM Reasoning (NeurIPS 2025 · similarity 0.87)
- DAG-Math: Graph-Guided Mathematical Reasoning in LLMs (ICLR 2026 · similarity 0.87)
- Monitoring the Internal Monologue: Probe Trajectories Reveal Reasoning Dynamics (NeurIPS 2026 · similarity 0.87)
- Reasoning on the Manifold: Bidirectional Consistency for Self-Verification in Diffusion Language Models (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.
