On the learnability and usage of acyclic probabilistic finite automata (Q1271548): Difference between revisions

The authors' main effort is to propose and analyze a distribution learning algorithm for certain acyclic probabilistic finite automata. Special attention is given to the modeling of short sequences that correspond to objects such as short DNA sequences, spoken words, or single handwritten letters. Specifically, they construct models for cursive handwritten letters (but no ideograms) and pronunciation models for spoken words. Formal results are given in the form of two theorems and several lemmas.