Marginalizing Out Transition Probabilities for Several Subclasses of PFAs

11th International Conference on Grammatical Inference

Chihiro Shibata [e-mail]

PID  :  120765

来源: CEUR

【 摘 要 】

A Bayesian manner which marginalizes transition probabilities can be generally applied to various kinds of probabilistic finite state machine models. Based on such a Bayesian man ner, we implemented and compared three algorithms: variablelength gram, state merging method for PDFAs, and collapsed Gibbs sampling for PFAs. Among those, collapsed Gibbs sampling for PFAs performed the best on the data from the precompetition stage of PAu tomaC, although it consumes large computation resources.