Research map / Graph learning / Graph algorithms and clustering
Failure-Driven Workflow Refinement
Jusheng Zhang, Kaitong Cai, Jing Yang, Ziliang Chen, Yongsen Zheng, Kwok-Yan Lam, Liang Lin, Keze Wang
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 structured operator library (ablation) | CE-Graph · MATH | accuracy: 47.35 -> 55.91 | Improved |
| add structured operator library (ablation) | CE-Graph · HumanEval | pass@1: 87.20 -> 94.26 | Improved |
| add verification step in propose-and-verify (ablation) | CE-Graph · MATH | accuracy: 49.10 -> 55.91 | Improved |
| add verification step in propose-and-verify (ablation) | CE-Graph · HumanEval | pass@1: 89.75 -> 94.26 | Improved |
| add failure clustering (ablation) | CE-Graph · MATH | accuracy: 51.25 -> 55.91 | Improved |
| add failure clustering (ablation) | CE-Graph · HumanEval | pass@1: 91.50 -> 94.26 | Improved |
| add convergence-aware stopping (ablation) | CE-Graph · MATH | optimization cost (10^3 tokens): 1,850 -> 1,210 | Improved |
| add convergence-aware stopping (ablation) | CE-Graph · MATH | accuracy: 53.50 -> 55.91 | Improved |
| add convergence-aware stopping (ablation) | CE-Graph · HumanEval | pass@1: 93.80 -> 94.26 | No clear effect |
| use BERT-base-uncased embeddings instead of SentenceTransformers (ablation) | CE-Graph · MATH | accuracy: 54.75 -> 55.91 | No clear effect |
| use BERT-base-uncased embeddings instead of text-embedding-ada-002 (ablation) | CE-Graph · MATH | accuracy: 55.34 -> 55.91 | No clear effect |
| replace MaAS with CE-Graph failure-driven refinement | CE-Graph vs MaAS · MBPP | pass@1: 82.17 -> 88.10 | Improved |
Closest papers by meaning
- Flow: Modularized Agentic Workflow Automation (ICLR 2025 · similarity 0.86)
- Benchmarking Agentic Workflow Generation (ICLR 2025 · similarity 0.86)
- GraphFlow: A Graph-Based Workflow Management for Efficient LLM-Agent Serving (ICML 2026 · similarity 0.86)
- Lean Refactor: Multi-Objective Controllable Proof Optimization via Agentic Strategy Search (NeurIPS 2026 · similarity 0.86)
- Automata from Agent Traces: Failure and Next-Step Prediction (NeurIPS 2026 · similarity 0.86)
- Optimize Once, Execute Fast: Latency-Aware Multi-Agent Workflow Learning for Recurrent Queries (NeurIPS 2026 · similarity 0.85)
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.
