Research map / Large language models / Agents, code and math
Textual Equilibrium Propagation for Deep Compound AI Systems
Minghui Chen, Wenlong Deng, James Zou, Han Yu, Xiaoxiao Li
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 |
|---|---|---|---|
| replace global textual backprop (TextGrad) with TEP local two-phase optimization | compound AI pipeline · HotpotQA | F1: 24.86 -> 48.72 | Improved |
| replace TextGrad with TEP | compound AI pipeline · BigCodeBench | pass@1: 35.71 -> 38.97 | Improved |
| add summarization to TextGrad feedback | TextGrad · HotpotQA | F1: 24.86 -> 24.12 | Hurt |
| add summarization to TextGrad feedback | TextGrad · STARK-PRIME | MRR: 41.31 -> 40.72 | Hurt |
| add the nudged phase to TEP (ablation) | TEP · HotpotQA | F1: 22.3 -> 48.72 | Improved |
| add the free phase to TEP (ablation) | TEP · HotpotQA | F1: -11.9 F1 without free phase | Improved |
| add the free phase to TEP (ablation) | TEP · BigCodeBench | pass@1: -2.7 pass@1 without free phase | Improved |
| apply TEP instead of TextGrad for solution optimization | Llama 3.2 · Object Counting | accuracy: 74.2 -> 81.6 | Improved |
| apply TEP instead of TextGrad for solution optimization | GPT-4 · GPQA | accuracy: 41.0 -> 44.5 | Improved |
| increase workflow depth with TextGrad (ablation) | TextGrad · BigCodeBench depth scaling | feedback tokens: 2K at s=1 -> over 32K at s=5 | Hurt |
| increase workflow depth with TextGrad w/ Sum (ablation) | TextGrad w/ Sum · BigCodeBench depth scaling | effective update rate: 36% -> 5% | Hurt |
| increase workflow depth with TEP (ablation) | TEP · BigCodeBench depth scaling | update success rate: 37% -> 33% | Mixed |
Closest papers by meaning
- AgentGrad: Intervention-guided Prompt Optimization for Multi Agent Systems (NeurIPS 2026 · similarity 0.87)
- Revolve: Optimizing AI Systems by Tracking Response Evolution in Textual Optimization (ICML 2025 · similarity 0.87)
- Recursive Introspection: Teaching Language Model Agents How to Self-Improve (NeurIPS 2024 · similarity 0.85)
- Retroformer: Retrospective Large Language Agents with Policy Gradient Optimization (ICLR 2024 · similarity 0.85)
- AgentTTS: Large Language Model Agent for Test-time Compute-optimal Scaling Strategy in Complex Tasks (NeurIPS 2025 · similarity 0.85)
- Test-Time Alignment for Large Language Models via Textual Model Predictive Control (ICLR 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.
