Inferring probabilistic acyclic automata using the minimum description length principle
Yaron Singer, Naftali Tishby · 2002
The use of Rissanen's (1978) minimum description length principle for the construction of probabilistic acyclic automata (PAA) is explored. We propose a learning algorithm for a PAA that is adaptive both in the structure and the dimension of the model. The proposed algorithm was tested on synthetic data as well as on real pattern recognition problems.>