Learning hidden Markov models with unknown number of states
Zheng, Jing 1Yu, Dongjie 1Zhu, Bin 1Tong, Changqing1
扫码查看
点击上方二维码区域，可以放大扫码查看
作者信息
1. Hangzhou Dianzi Univ
折叠
Abstract
This paper proposes a novel method for learning hidden Markov models (HMMs) with an unknown number of states based on a valuable feature set. The feature set is built using the hitting times of HMMs. Based on the feature set, we obtain a parameter estimation for HMMs by SVD and the clustering algorithm. The advantages of the proposed method are that it can accurately and automatically identify the number of hidden states, it is robust to misspecified emission distributions, it is less sensitive to initialization, and is asymptotically consistent. Numerical experiments show that the proposed method performs better than other methods when the observed time series is long enough. (c) 2022 Elsevier B.V. All rights reserved.
Key words
Hidden Markov model/Spectral clustering algorithm/The first hitting time/INFERENCE
NSTL
Elsevier