Pattern Recognition in Sequences Using Multistate Sequence Autoencoding Neural Networks

In the paper, we describe the Multistate Sequence Autoencoding Neural Networks (MSANNs) that enable us to deal with some problems concerning pattern recognition in sequential data. The properties of the MSANN networks proposed by us make these networks suitable for use in the following situations: patterns appear only partially, patterns are in different orders, patterns are stretched over time, etc. The training and testing procedures for the MSANN networks are presented in the paper.