Research map / Large language models / Agents, code and math
CAMA: Enhancing Mathematical Reasoning in Large Language Models with Causal Knowledge
Lei Zan, Keli Zhang, Ruichu Cai, Lujia Pan
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 MCG causal-graph guidance (CAMA) vs CoT zero-shot | DeepSeek-V3-0324 · AIME2024 | Pass@1: 42.2% -> 50.0% | Improved |
| add MCG guidance vs CoT zero-shot | DeepSeek-V3-0324 · OlympiadBench-674 | Pass@1: 67.2% -> 66.4% | No clear effect |
| add MCG guidance vs CoT zero-shot | Qwen3-32B · AIME2024 | Pass@1: 75.6% -> 82.2% | Improved |
| few-shot CoT examples instead of zero-shot | Qwen3-32B · AIME2025 | Pass@1: 72.2% -> 61.1% | Hurt |
| use directed edges instead of undirected (ablation) | CAMA (DeepSeek-V3) · AIME2024 | Pass@1: 43.3% -> 50.0% | Improved |
| use directed edges instead of undirected (ablation) | CAMA (Qwen3-32B) · AIME2025 | Pass@1: 72.2% -> 76.7% | Improved |
| add alignment stage (ablation) | CAMA (DeepSeek-V3) · AIME2025 | Pass@1: 38.9% -> 38.9% | No clear effect |
| add alignment stage (ablation) | CAMA (Qwen3-32B) · AIME2024 | Pass@1: 78.9% -> 82.2% | Improved |
| add alignment stage (ablation) | CAMA (DeepSeek-V3) · OlympiadBench-674 | Pass@1: 66.8% -> 66.4% | Hurt |
| increase knowledge point granularity lambda (ablation) | CAMA (DeepSeek-V3) · AIME2024 (test) | Pass@1 | Hurt |
Closest papers by meaning
- Unveiling Causal Reasoning in Large Language Models: Reality or Mirage? (NeurIPS 2024 · similarity 0.89)
- COLD: Causal reasOning in cLosed Daily activities (NeurIPS 2024 · similarity 0.89)
- Compositional Causal Reasoning Evaluation in Language Models (ICML 2025 · similarity 0.88)
- Selection, Reflection and Self-Refinement: Revisit Reasoning Tasks via a Causal Lens (ICLR 2026 · similarity 0.88)
- MARGE: Improving Math Reasoning with Guided Exploration (ICML 2025 · similarity 0.88)
- Mitigating Hallucinations in Large Language Models via Causal Reasoning (AAAI 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.
