Zero Isn’t a Problem, It’s a Shortcut: Rethinking PDFA Learning

8 Apr 2025

We propose a fast learning algorithm for quotient PDFA to model LLMs under constrained generation and validate its efficiency and real-world utility.

What Happens When Language Models Say 'Nothing'?

8 Apr 2025

We propose a new congruence and learning algorithm to analyze constrained neural text generation using quotient probabilistic automata.