中央研究院 資訊科學研究所

活動訊息

友善列印

列印可使用瀏覽器提供的(Ctrl+P)功能

Extensions of hidden Markov models for variable frame rate and piecewise linear random sequences.

:::

Extensions of hidden Markov models for variable frame rate and piecewise linear random sequences.

  • 講者李立民 教授 (Department of Electrical Engineering, Da-Yeh University)
    邀請人:王新民
  • 時間2016-07-12 (Tue.) 10:30 ~ 12:30
  • 地點資訊所新館101演講廳
摘要

In this talk, we will present two extensions of hidden Markov models (HMMs), one for variable frame rate processes and the other for piecewise linear random sequences. A traditional HMM generates piecewise stationary random sequences and is often used to model fixed frame rate sequences. However, a random sequence in real world may be better fit by a piecewise linear model and the frame rate of the sequence may be variable.

We will first discuss the transformation of HMM for variable frame rate sequences. A variable frame rate sequence can be regarded as a sequence resulted from irregular losing of data. Therefore, the transition probabilities of the HMM should reflect the losing of data and the resulted model is an HMM with dynamic transition probabilities. Then a discussion on high-order HMM (HO-HMM) and its extension to the modeling of piecewise linear random sequences is presented. In the proposed piecewise linear HO-HMM, the mean vector of a state in an HO-HMM is a linear function of time instead of a fixed vector. We shall derive recursive formulas for the forward and backward functions and the expectation-maximization algorithm for the learning of such models. Because the mean vector of a state at different duration is linearly related instead of freely positioned, the parameters for a high-order state at different durations are jointly determined from all the data aligned to that state.

BIO

Lee-Min Lee graduated from National Taipei Institute of Technology in 1982 and received the M.S. and Ph.D. degrees from National Tsing Hua University, Hsinchu, Taiwan in 1989 and 1995, respectively, all in electrical engineering. From 1995 to 2002, he joined the faculty of Mingchi Institute of Technology. He is now a professor with the Department of Electrical Engineering of Da-Yeh University, Changhua, Taiwan. His research interests include speech recognition, machine learning, and natural language processing.