Research map / Large language models / Agents, code and math
Multi-View Encoders for Performance Prediction in LLM-Based Agentic Workflows
Patara Trirat, Wonyong Jeong, Sung Ju Hwang
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 |
|---|---|---|---|
| use multi-view (code+graph+text) encoder with pretraining-free supervised training vs best GNN baseline | workflow performance predictor · FLORA-Bench average | accuracy: 78.36 -> 79.97 | Improved |
| same, utility | workflow performance predictor · FLORA-Bench average | utility: 73.54 -> 76.33 | Improved |
| Agentic Predictor vs GCN on reasoning utility | workflow performance predictor · FLORA-Bench ReasonAF | utility: 91.82 -> 91.47 | Hurt |
| add code view to graph+text views (ablation) | Agentic Predictor · AFlow average | accuracy: 75.87 -> 84.38 | Improved |
| add text view to code+graph views (ablation) | Agentic Predictor · AFlow average | accuracy: 82.15 -> 84.38 | Improved |
| add graph view to code+text views (ablation) | Agentic Predictor · AFlow average | accuracy: 70.82 -> 84.38 | Improved |
| graph-only vs full three views on reasoning utility (ablation) | Agentic Predictor · AFlow Common Reasoning | utility: 91.51 -> 91.47 | Hurt |
| use multiple graphs instead of single graph (ablation) | Agentic Predictor · AFlow code generation | accuracy: 82.58 -> 84.44 | Improved |
| use multiple graphs instead of single graph (ablation) | Agentic Predictor · AFlow code generation | utility: 78.52 -> 77.22 | Hurt |
| add cross-domain unsupervised pretraining at low label ratio | Agentic Predictor+ · label ratio 0.1 | accuracy: ~70 -> >73 | Improved |
| use graph predictor instead of 5-shot LLM predictor | Agentic Predictor · FLORA-Bench | accuracy: 62.86 -> 84.97 | Improved |
| train on AFlow and test on G-Designer (OOD) | Agentic Predictor · cross-system OOD | accuracy: 59.52 -> 62.05 | Improved |
Closest papers by meaning
- Optimize Once, Execute Fast: Latency-Aware Multi-Agent Workflow Learning for Recurrent Queries (NeurIPS 2026 · similarity 0.90)
- Benchmarking Agentic Workflow Generation (ICLR 2025 · similarity 0.89)
- AgentSquare: Automatic LLM Agent Search in Modular Design Space (ICLR 2025 · similarity 0.89)
- Generalizable End-to-End Tool-Use RL with Synthetic CodeGym (ICLR 2026 · similarity 0.88)
- GraphFlow: A Graph-Based Workflow Management for Efficient LLM-Agent Serving (ICML 2026 · similarity 0.88)
- AgentBench: Evaluating LLMs as Agents (ICLR 2024 · 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.
