Research map / Large language models / Agents, code and math
A²Flow: Automating Agentic Workflow Generation via Self-Adaptive Abstraction Operators
Mingming Zhao, Xiaokang Wei, Yuanqi Shao, Kaiwen Zhou, Lin Yang, Siwei Rao, Junhui Zhan, Zhitang Chen
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 Operators Memory (ablation) | A2Flow · MATH (level 5) | score: 53.9 -> 58.5 | Improved |
| add case-based initial operator generation (ablation) | A2Flow · MATH (level 5) | score: 49.6 -> 58.5 | Improved |
| add operator clustering (ablation) | A2Flow · MATH (level 5) | score: 54.5 -> 58.5 | Improved |
| add deep extraction of abstract execution operators (ablation) | A2Flow · MATH (level 5) | score: 51.6 -> 58.5 | Improved |
| use abstract operators without initial operators or without deep extraction, versus no abstraction operators and no memory (ablation) | A2Flow · MATH (level 5) | score: 56.2 -> 49.6 (no initial), 51.6 (no deep extraction) | Hurt |
| replace AFLOW predefined operators with A2Flow abstract operators | agentic workflow search · DROP | F1: 80.6 -> 85.1 | Improved |
| replace AFLOW predefined operators with A2Flow abstract operators | agentic workflow search · MATH (level 5) | solve rate: 56.2 -> 58.5 | Improved |
| replace AFLOW predefined operators with A2Flow abstract operators | agentic workflow search · HumanEval | pass@1: 90.9 -> 92.4 | Mixed |
| replace AFLOW predefined operators with A2Flow abstract operators | agentic workflow search · GSM8K | solve rate: 93.5 -> 93.8 | Improved |
| use A2Flow instead of AFLOW on embodied tasks | agentic workflow search · ALFWorld seen | success rate: 17.1 -> 25.0 | Improved |
| use A2Flow instead of AFLOW on TextCraft | agentic workflow search · TextCraft | success rate: 53.0 -> 59.0 | Improved |
| use ADAS instead of manual baselines | automated workflow search · MBPP | pass@1: IO 71.8 -> ADAS 53.4 | Hurt |
Closest papers by meaning
- DyFlow: Dynamic Workflow Framework for Agentic Reasoning (NeurIPS 2025 · similarity 0.92)
- AFlow: Automating Agentic Workflow Generation (ICLR 2025 · similarity 0.90)
- GraphFlow: A Graph-Based Workflow Management for Efficient LLM-Agent Serving (ICML 2026 · similarity 0.90)
- Flow: Modularized Agentic Workflow Automation (ICLR 2025 · similarity 0.90)
- Benchmarking Agentic Workflow Generation (ICLR 2025 · similarity 0.90)
- Optimize Once, Execute Fast: Latency-Aware Multi-Agent Workflow Learning for Recurrent Queries (NeurIPS 2026 · similarity 0.89)
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.
