Deep learning anomaly detection is an evolving field with many real-world applications. As more and more devices continue to be added to the Internet of Things (IoT), there is an increasing desire to make use of the additional computational capacity to run demanding tasks. The increase in devices and amounts of data flooding in have led to a greater need for security and outlier detection. Motivated by those facts, this thesis studies the potential of creating a distributed anomaly detection framework. While there have been vast amounts of research into deep anomaly detection, there has been no research into building such a model in a distributed context. In this work, we propose an implementation of a distributed anomaly detection system using the TensorFlow library in Python and three Nvidia Jetson AGX Xavier deep learning modules. The key objective of this study is to determine if it is practical to create a distributed anomaly detection model without a significant loss of accuracy on classification. We then present an analysis of the performance of the distributed system in terms of accuracy and runtime and compare it to a similar system designed to run on a single device. The results of this study show that it is possible to build a distributed anomaly detection system without a significant loss of accuracy using TensorFlow, but the overall runtime increases for these trials. This proves that it is possible to distribute anomaly detection to edge devices without sacrificing accuracy, and the runtime can be improved with further research.
If this is your thesis or dissertation, and want to learn how to access it or for more information about readership statistics, contact us at STARS@ucf.edu
Master of Science (M.S.)
College of Engineering and Computer Science
Length of Campus-only Access
Masters Thesis (Open Access)
Holdren, William, "Deep Learning Anomaly Detection Using Edge AI" (2022). Electronic Theses and Dissertations, 2020-. 1026.