Skip to Main Content
Hidden Markov model (HMM) is a highly effective mean of modeling a common motif within a set of unaligned sequences, which has been proved to be a prior tool in similarity search analysis based on time-series data . However, its major drawback is that its training process is computationally expensive, which makes it hard to be efficient and precise simultaneously. In this paper, an efficient HMM-based similarity search scheme is proposed with an innovative training algorithm using small size of training data composed of only distinct subsequences, which is very useful for the metamaterial design. Experiment results show that the training time of our method can be reduced extremely to 1% of that of conventional methods. Furthermore, our HMM-based model is more stable with threshold fluctuating, which make it more feasible in practice.