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
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 |
|---|---|---|---|
| use CAPAL class query + discrimination tree instead of LearnAnyway | LearnAnyway · DFA learning, Code-based MQ oracle | fraction of DFAs learned exactly: 0.786 -> 0.928 | Improved |
| use CAPAL instead of LearnAnyway | LearnAnyway · DFA learning, Discriminator & CoT oracle | fraction of DFAs learned exactly: 0.714 -> 0.821 | Improved |
| use CAPAL instead of LearnAnyway | LearnAnyway · DFA learning, Verification & CoT oracle | fraction of DFAs learned exactly: 0.714 -> 0.679 | No clear effect |
| use CAPAL instead of TTT/L* under noisy baseline prompts | TTT / L* · DFA learning, Baseline oracle | fraction of DFAs learned exactly: 0.107 -> 0.250 (TTT) | Improved |
| replace baseline prompt with chain-of-thought | MQ oracle · DFA learning | MQ error rate: 0.156 -> 0.068 | Improved |
| add verification step to CoT | MQ oracle · DFA learning | MQ error rate: 0.156 -> 0.044 | Improved |
| add discriminator-guided hints to CoT | MQ oracle · DFA learning | MQ error rate: 0.156 -> 0.047 | Improved |
| replace prompt-based oracle with LLM-synthesized code oracle | MQ oracle · DFA learning | MQ error rate: 0.156 -> 0.013 | Improved |
| use CoT prompting with CAPAL | CAPAL · DFA learning | LLM calls per DFA: 43.75 -> 30.93 | Improved |
| use CoT prompting with LearnAnyway | LearnAnyway · DFA learning | LLM calls per DFA: 46.07 -> 52.64 | Hurt |
| use majority-vote resampling of persistent-noise MQs | MQ oracle · pMAT membership queries | error of aggregated label | Hurt |
| use CAPAL instead of passive RPNI-EDSM at high noise (eps > 0.18) | CAPAL · DFA learning, noise sweep | MQ efficiency | Mixed |
Closest papers by meaning
- From Passive to Active Reasoning: Can Large Language Models Ask the Right Questions under Incomplete Information? (ICML 2025 · similarity 0.87)
- Efficient Evaluation of LLM Performance with Statistical Guarantees (NeurIPS 2026 · similarity 0.87)
- Efficiently Learning at Test-Time: Active Fine-Tuning of LLMs (ICLR 2025 · similarity 0.87)
- Conformal Information Pursuit for Interactively Guiding Large Language Models (NeurIPS 2025 · similarity 0.87)
- Soft Reasoning: Navigating Solution Spaces in Large Language Models through Controlled Embedding Exploration (ICML 2025 · similarity 0.87)
- Sample Smart, Not Hard: Correctness-First Decoding for Better Reasoning in LLMs (ICLR 2026 · similarity 0.87)
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.
