Year: 2018
Type: article
Abstract: In this dissertation, we propose a new sequential clustering method based on nonlinear dynamic theories and exponential random graph models (ERGMs), a class of social network models. In particular, we convert each sequence to its recurrence network and make use of the flexibility and information of the network. Our algorithm shows good performance on simulated data, real-world data and public benchmark data based on clustering evaluation metrics. To make sure the connection between a sequence and a network is reliable, we also conduct a study to examine whether the network contains enough information of the original sequence. We consider recurrence networks as images and build state of the art convolutional neural network (CNN) models based on image data to predict the labels of original sequences. Our method is very competitive compared with other advanced sequential classification methods. This also verifies that recurrence networks are very informative and give us enough information for real-world applications.
Author: Zhang Zhang
Language: English
Subfield: Signal Processing
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2803175399