AI Paper Atlas

Research map / Large language models / Agents, code and math

Evolving Graph Structured Programs for Circuit Generation with Large Language Models

Yinqi Bai, Jie Wang, Lei Chen, Zhihai Wang, Yumeng Li, Mingxuan Yuan, Jianye HAO, Defu Lian, Enhong Chen

ICLR 2026 · Large language models › Agents, code and math · approach: Software engineering agents

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
use an LLM generator instead of a rule-based linear genetic program (ablation)CircuitEvo (GPT-3.5-turbo) · Espresso4init nodes: 940 -> 848Improved
use the graph-structured circuit program instead of a Boolean function format (ablation)CircuitEvo (GPT-3.5-turbo) · Espresso4init nodes: 940 -> 848Improved
add exploration and refinement evolutionary strategies (ablation)CircuitEvo (GPT-3.5-turbo) · Espresso4init nodes: 911 -> 848Improved
add local program search before function completion (ablation)CircuitEvo (GPT-3.5-turbo) · Espresso4init nodes: 911 -> 848Improved
add function completion (ablation)CircuitEvo (GPT-3.5-turbo) · Espresso4accuracy (%): 86.4 -> 100.0Improved
add function completion (ablation)CircuitEvo (GPT-3.5-turbo) · Arithmetic3accuracy (%): 67.8 -> 100.0Improved
add local program search before function completion (ablation)CircuitEvo (GPT-3.5-turbo) · Arithmetic1init nodes: 79 -> 74Improved
use an LLM generator instead of a rule-based linear genetic program (ablation)CircuitEvo (GPT-3.5-turbo) · LogicNets2init nodes: 90 -> 74Improved
use the graph-structured circuit program instead of a Boolean function format (ablation)CircuitEvo (GPT-3.5-turbo) · LogicNets2init nodes: 74 -> 74No clear effect
add evolutionary strategies (ablation)CircuitEvo (GPT-3.5-turbo) · Arithmetic3init nodes: 1233 -> 1206Improved
use CircuitEvo (Deepseek-V3) instead of DNASDNAS · Initial circuit size, average of 16 circuitsinit nodes: 532.88 -> 470.13Improved
use CircuitEvo instead of DNASDNAS · Convergence time, average of 4 circuitstime (h): 16.91 -> 1.48Improved

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