Dynamic Models for Sequential Data Analysis
Arash Gharehbaghi · 2023
Time series variations in time, so called time series dynamics, can convey important information, not only about the time series by itself, but also about the system behind, which generates the time series. Exploring dynamics of time series is sometimes performed directly by using the time series values only. In many other situations, a dynamic generative model is hyphenized and the learning process is performed based on the model parameters. This chapter begins with a description of dynamic time warping method, known as a method which considers merely time series values, and performs the learning task according to the time series structure. Various forms of this method is described. Hidden Markov model, is detailed in the next section, as a model-based method to learn the generative model using statistical methods. The learning task is however performed based on the model parameters. This chapter ends with an introduction about recurrent neural network, which is a model-based dynamic learning method. The model parameters are numerically found in an iterative way. This last section is only an introductory section and the details are postponed until Chapter 11.