AI Paper Atlas

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

AnalogVerifier: A Neuro-Symbolic Framework for Analog Circuit Verification

Yanfang Liu, Mingjun Wang, Peng XU, Rongliang Fu, Bei Yu, Tsung-Yi Ho

ICML 2026 · Large language models › Agents, code and math · approach: Inductive and abstract reasoning evaluation

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
add Closed-Loop Repair (ablation)AnalogVerifier · OTA / HSCpass rate: 33.3 -> 92.3 (OTA); 50.6 -> 95.1 (HSC)Improved
add Numerical-Symbolic Grounding (cumulative removal) (ablation)AnalogVerifier · OTA / HSCpass rate: 33.3 -> 23.1 (OTA); 50.6 -> 37.0 (HSC) when removedImproved
add Graph-Symbolic Scheduling (cumulative) (ablation)AnalogVerifier · OTA / HSCpass rate: 23.1 -> 13.1 (OTA); 37.0 -> 20.6 (HSC)Improved
add Context-Aware Serialization (cumulative) (ablation)AnalogVerifier · OTA / HSCpass rate: 13.1 -> 7.2 (OTA); 20.6 -> 11.3 (HSC)Improved
replace Numerical-Symbolic Grounding with LLM fallback (ablation)AnalogVerifier · OTA / HSC independentpass rate: 92.3 -> 53.8 (OTA); 95.1 -> 55.6 (HSC)Hurt
replace Graph-Symbolic Scheduling with LLM fallback (ablation)AnalogVerifier · OTA / HSC independentpass rate: 92.3 -> 61.5 (OTA); 95.1 -> 59.3 (HSC)Hurt
replace Context-Aware Serialization with LLM fallback (ablation)AnalogVerifier · OTA / HSC independentpass rate: 92.3 -> 54.2 (OTA); 95.1 -> 57.8 (HSC)Hurt
add the full AnalogVerifier pipeline to direct LLM testbench generationLLM backbones · five industrial analog circuitspass rate: up to 9.1 -> 82.3%-100.0%Improved
add structured serialization, scheduling and grounding (no repair)LLM backbones · five analog circuitsextracted tasks: 2-7x more tasksImproved
use Claude-Haiku-4.5 instead of GPT-5.1AnalogVerifier w/o Repair · extracted tasks vs precisionprecision: e.g. HSC 118 tasks, 38.8 precision vs 44 tasks, 46.5Mixed
add closed-loop repair after generationAnalogVerifier · OTA case study (Claude-Haiku-4.5)first-pass pass rate: 69.0% -> 96.4%Improved

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