Quaternion Temporal Convolutional Neural Networks

Date of Award


Degree Name

M.S. in Computer Engineering


Department of Electrical and Computer Engineering


Advisor: Vijayan Asari


Sequence Processing and Modeling are a domain of problems recently receiving significant attention for significant advancements in research and technology. While traditionally sequence processing using neural networks has been done using a recurrent neural network such as the long-short term memory cell. These recurrent networks have some fairly large drawbacks. issue in networks is increasingly large networks, which have been proven to learn features from useless noise in their input data. A network called the Temporal Convolutional Network seeks to fix the issues that the long-short term memory cell have. While other recent research has been put into quaternion neural networks, networks that dramatically reduce the number of parameters in a network while keeping the same performance. This thesis combines both these recent advancements into a Quaternion Temporal Convolutional Network. The network performance is evaluated on a wide range of sequence processing and modeling tasks and compared to the base Temporal Convolutional Network. Through testing and evaluation it is shown that although there is a reduction in the number of learned parameters in the Temporal Convolutional network by up to 4x, the network performance stays relatively close, and actually beats the base network on some tasks.


Computer Science, Engineering, Quaternion Temporal Convolutional Network, QTCN, Quaternion Neural Network, Sequence Processing, Machine Learning

Rights Statement

Copyright 2019, author