情報工学科の大学生が学ぶこと

今学んでいることが何に結びつくのかを理解し、効率的に実践的な知識を身につける術を提供します。

隠れマルコフモデル(HMM)における基本的な3つのアルゴリズム

観測した単語列の確率を推定

Estimate the probability of an observed sequence. This corresponds to the sum of all the paths producing the observation. It is solved using the forward procedure.
In the specific case of POS tagging, it will determine the probability of the word sequence. Although the forward procedure is not of primary importance here, it is fundamental and has many other applications.

観測された単語列のもっとも確率が高いパスを決定

Determine the most likely path of an observed sequence. This is a decoding problem that is solved using the Viterbi algorithm.

観測した学習データからパラメータを算出

Determine (learn) the parameters given a set of observations. This algorithm is used to build models when we do not know the parameters. It is solved using the forward–backward algorithm.

参考

A tutorial on hidden Markov models and selected applications in speech recognition - Proceedings of the IEEE

Language Processing With Perl and Prolog: Theories, Implementation, and Application (Cognitive Technologies)

Language Processing With Perl and Prolog: Theories, Implementation, and Application (Cognitive Technologies)