隠れ状態の継続時間長を考慮した確率モデルに関する調査

抄録

Human activity data, so-called "life-log data", has underlying contexts such as sleeping, driving a car, eating. In order to analyse and predict such data, the hidden Markov model is often used. However, the duration time of the hidden context state of HMM distributes exponentially and this is not suited for modeling the above contexts. In order for modeling these context flexibly, we introduce and compare probabilistic modeling techniques of of hidden states with general duration distributions.