Degree

Doctor of Philosophy (PhD)

Department

Electrical Engineering

Document Type

Dissertation

Abstract

In the last decade, techniques for artificial intelligence (AI) has advanced tremendously, which lead to solutions to many problems that have long-troubled us. Such examples include image/video recognition, speech recognition, and 3D scenario recognition. As the tool become more and more powerful, we started to explore data types that have never been handled in an AI fashion. Graph is undoubtedly the first one that comes to mind. Many important real-life data is or can be represented as graphs or networks: social networks, communication networks, protein-protein interaction networks, molecular structures, etc. Yet very little attention has been devoted to the study of the graph information processing in terms of AI systems until the very recent few years. For centuries people have used mathematics to solve graph problems and it worked really well until recent decades. With the rapid development of Internet, the information/data available to us has grown exponentially. It makes it very difficult for conventional mathematic tools to solve new graph problem since it will just take too much time to calculate. Thus, people turned their attention to the newborn techniques (convolutional neural network (CNN), recurrent neural network (RNN), reinforcement learning (RL), etc.) of AI. And once again, AI has shown its power to us. In this work, I will present some techniques of graph information processing for AI. The work will concentrate on two different aspect of graph information processing: information processing of non-graph data and graph data. In the first part, we started with non-graph data, signals in our case. Then a conversion is employed to convert the data into graph based on the objective and nature of the data. In the second part, methods to aggregate information in graph will be illustrated. The whole report contains three researches: tag recognition of radio frequency identification (RFID) in Internet of Things (IoT), photoplethysmogram (PPG) signal based authentication system, and cancer target prediction.

Committee Chair

Wu, Hsiao-Chun

DOI

10.31390/gradschool_dissertations.4957

Download

Share

COinS