A probablistic automata learning approach for analyzing and sampling constrained LLM

Matías Carrasco, Franz Mayr, Sergio Yovine, Johny Kidd, Martin Iturbide, Juan Pedro da Silva Barloco, Alejo Garat

Keywords: Grammatical Inference, Probabilistic Deterministic Finite Automata, Active Learning, LLM

TL;DR: We develop an algorithm for efficiently learning the quotient PDFA with respect to a congruence that copes with null next-symbol probabilities that arise when the output of a language model is constrained

Abstract: We define a congruence that copes with null next-symbol probabilities that arise when the output of a language model is constrained by some means during text generation. We develop an algorithm for efficiently learning the quotient with respect to this congruence and evaluate it on case studies for analyzing statistical properties of LLM.

Primary Area: neurosymbolic & hybrid AI systems (physics-informed, logic & formal reasoning, etc.)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 8404