Discriminative Learning in the Model Space for Symbolic Sequence Classification

Yaqiang YAO, Huanhuan CHEN, Xin YAO

Research output: Journal PublicationsJournal Article (refereed)peer-review

Abstract

Learning in the model space is an approach that represents sequences with state space models and performs the learning procedure in the space spanned by the “model points.” To enhance the discriminative capacity and preserve the simplicity and effectiveness of model-based kernel, a discriminative learning method in the model space is proposed for symbolic sequence classification in this paper. Compared with string kernels, the discriminative model-based kernel not only models the temporal information of sequence by a generative mechanism but also considers label information of sequence in the similarity measurement, which is beneficial to sequence classification. On the other hand, although learning in the model space approach has been successfully applied in the cognitive fault diagnosis and the time-series analysis, it is actually unclear how it might perform on the symbolic sequence classification. This paper extends the learning in the model space approach to the symbolic sequence classification problem with a coding scheme on categorical variables. Experimental results and the related analysis confirm the effectiveness of discriminative learning in the model space approach for the symbolic sequence classification. Moreover, we investigate several characteristics related to the performance of the proposed approach, including parameter analysis, Lyapunov analysis etc., and present a guideline on exploiting the discriminative learning in the model space approach in real-world applications.
Original languageEnglish
Pages (from-to)154-167
Number of pages14
JournalIEEE Transactions on Emerging Topics in Computational Intelligence
Volume5
Issue number2
DOIs
Publication statusPublished - Apr 2021
Externally publishedYes

Keywords

  • Learning in the model space
  • sequence classification
  • reservoir computing
  • discriminative learning

Fingerprint

Dive into the research topics of 'Discriminative Learning in the Model Space for Symbolic Sequence Classification'. Together they form a unique fingerprint.

Cite this