Research map / Multi-agent systems and games / LLM agents
Failure Localization in Multi-Agent Code Generation via Knowledge-Guided and Transferable Reasoning
Mingyang Geng, Shanzhi Gu, Zhipeng Liu, Chuanfu Xu, Zhaoyang Qu, Haotian Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Addresses failure localization in multi-agent LLM code generation using knowledge-guided and transferable reasoning (abstract only available).
Problem
Locating failures in multi-agent LLM code generation is hard due to inter-agent dependencies and multiple valid solution paths; existing prompting-based localization methods are vulnerable to semantically valid but non-canonical strategies.
Approach
Knowledge-guided and transferable reasoning for failure localization in multi-agent code generation; specific mechanism not stated in the available abstract.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Knowledge Projection Module (ablation) | FLKR · COFL general cases | Top-1 LA (%): 58 -> 78 | Improved |
| add contextual divergence score (ablation) | FLKR · COFL general cases | Top-1 LA (%): 64 -> 78 | Improved |
| add knowledge divergence score (ablation) | FLKR · COFL general cases | Top-1 LA (%): 57 -> 78 | Improved |
| add InfoNCE contrastive alignment loss (ablation) | FLKR · COFL general cases | Top-1 LA (%): 60 -> 78 | Improved |
| add self-distillation loss (ablation) | FLKR · COFL general cases | Top-1 LA (%): 70 -> 78 | Improved |
| add Knowledge Projection Module (ablation) | FLKR · COFL divergent-solution cases | Top-1 LA (%): 43 -> 62 | Improved |
| add self-distillation loss (ablation) | FLKR · COFL divergent-solution cases | Top-1 LA (%): 52 -> 62 | Improved |
| replace prompting-based localization with FLKR | Binary Search Prompting · COFL general cases | Top-1 LA (%): 35 -> 78 | Improved |
| replace prompting-based localization with FLKR | Binary Search Prompting · SoftwareDev out-of-domain | Top-1 LA (%): 30 -> 76 | Improved |
| increase training data (ablation) | FLKR · COFL | FLA / Top-1 LA | Improved |
| set distillation weight lambda near 0.7 and temperature 0.07-0.08 (ablation) | FLKR · COFL | FLA / Top-1 LA | Improved |
Limitations
- Full text unavailable; details of method and results unknown
Experimental setup
Baselines: prompting-based localization methods
Closest papers by meaning
- OrcaLoca: An LLM Agent Framework for Software Issue Localization (ICML 2025 · similarity 0.86)
- CORRECT: COndensed eRror RECognition via knowledge Transfer in multi-agent systems (ICML 2026 · similarity 0.86)
- DoVer: Intervention-Driven Auto Debugging for LLM Multi-Agent Systems (ICLR 2026 · similarity 0.85)
- Improving Code Localization with Repository Memory (ICLR 2026 · similarity 0.85)
- Which Agent Causes Task Failures and When? On Automated Failure Attribution of LLM Multi-Agent Systems (ICML 2025 · similarity 0.85)
- GitTaskBench: A Benchmark for Code Agents Solving Real-World Tasks Through Code Repository Leveraging (AAAI 2026 · similarity 0.84)
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.
