AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Towards Persistent Noise-Tolerant Active Learning of Regular Languages with Class Query

Lekai Chen, Ashutosh Trivedi, Alvaro Velasquez

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Statistical frameworks for LLM 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
use CAPAL class query + discrimination tree instead of LearnAnywayLearnAnyway · DFA learning, Code-based MQ oraclefraction of DFAs learned exactly: 0.786 -> 0.928Improved
use CAPAL instead of LearnAnywayLearnAnyway · DFA learning, Discriminator & CoT oraclefraction of DFAs learned exactly: 0.714 -> 0.821Improved
use CAPAL instead of LearnAnywayLearnAnyway · DFA learning, Verification & CoT oraclefraction of DFAs learned exactly: 0.714 -> 0.679No clear effect
use CAPAL instead of TTT/L* under noisy baseline promptsTTT / L* · DFA learning, Baseline oraclefraction of DFAs learned exactly: 0.107 -> 0.250 (TTT)Improved
replace baseline prompt with chain-of-thoughtMQ oracle · DFA learningMQ error rate: 0.156 -> 0.068Improved
add verification step to CoTMQ oracle · DFA learningMQ error rate: 0.156 -> 0.044Improved
add discriminator-guided hints to CoTMQ oracle · DFA learningMQ error rate: 0.156 -> 0.047Improved
replace prompt-based oracle with LLM-synthesized code oracleMQ oracle · DFA learningMQ error rate: 0.156 -> 0.013Improved
use CoT prompting with CAPALCAPAL · DFA learningLLM calls per DFA: 43.75 -> 30.93Improved
use CoT prompting with LearnAnywayLearnAnyway · DFA learningLLM calls per DFA: 46.07 -> 52.64Hurt
use majority-vote resampling of persistent-noise MQsMQ oracle · pMAT membership querieserror of aggregated labelHurt
use CAPAL instead of passive RPNI-EDSM at high noise (eps > 0.18)CAPAL · DFA learning, noise sweepMQ efficiencyMixed

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