作者: Yoshinori Sagisaka , Hideharu Nakajima , Sabine Deligne
DOI:
关键词: Sequence 、 Variable length 、 Probability distribution 、 Class (philosophy) 、 String (computer science) 、 Algorithm 、 Artificial intelligence 、 Mathematics 、 Bigram 、 Expectation–maximization algorithm 、 Pattern recognition 、 Sequence model
摘要: An apparatus generates a statistical class sequence model called A bi-multigram from input training strings of discrete-valued units, where bigram dependencies are assumed between adjacent variable length sequences maximum N and labels assigned to the sequences. The number times all units occur counted, as well pairs co-occur in strings. initial probability distribution is computed two co-occur, divided by first occurs string. Then, classified into pre-specified desired classes. Further, an estimate calculated using EM algorithm maximize likelihood string with distributions. above processes then iteratively performed generate model.