AI Paper Atlas

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

ICLR 2026 · Large language models › Agents, code and math · approach: Long-horizon LLM agents · 1 citations (OpenAlex)

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 changedWhereMeasuredEffect
replace global textual backprop (TextGrad) with TEP local two-phase optimizationcompound AI pipeline · HotpotQAF1: 24.86 -> 48.72Improved
replace TextGrad with TEPcompound AI pipeline · BigCodeBenchpass@1: 35.71 -> 38.97Improved
add summarization to TextGrad feedbackTextGrad · HotpotQAF1: 24.86 -> 24.12Hurt
add summarization to TextGrad feedbackTextGrad · STARK-PRIMEMRR: 41.31 -> 40.72Hurt
add the nudged phase to TEP (ablation)TEP · HotpotQAF1: 22.3 -> 48.72Improved
add the free phase to TEP (ablation)TEP · HotpotQAF1: -11.9 F1 without free phaseImproved
add the free phase to TEP (ablation)TEP · BigCodeBenchpass@1: -2.7 pass@1 without free phaseImproved
apply TEP instead of TextGrad for solution optimizationLlama 3.2 · Object Countingaccuracy: 74.2 -> 81.6Improved
apply TEP instead of TextGrad for solution optimizationGPT-4 · GPQAaccuracy: 41.0 -> 44.5Improved
increase workflow depth with TextGrad (ablation)TextGrad · BigCodeBench depth scalingfeedback tokens: 2K at s=1 -> over 32K at s=5Hurt
increase workflow depth with TextGrad w/ Sum (ablation)TextGrad w/ Sum · BigCodeBench depth scalingeffective update rate: 36% -> 5%Hurt
increase workflow depth with TEP (ablation)TEP · BigCodeBench depth scalingupdate success rate: 37% -> 33%Mixed

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy