Abstract

With data explosion in many domains, such as social media, big code repository, Internet of Things (IoT), and inertial sensors, only 32% of data available to academic and industry is put to work, and the remaining 68% goes unleveraged. Moreover, people are facing an increasing number of obstacles concerning complex analytics on the sheer size of data, which include 1) how to perform dynamic graph analytics in a parallel and robust manner within a reasonable time? 2) How to conduct performance optimizations on a property graph representing and consisting of the semantics of code, data, and runtime systems for big data applications? 3) How to innovate neural graph approaches (ie, Transformer) to solve realistic research problems, such as automated program repair and inertial navigation? To tackle these problems, I present two efforts along this road: efficient graph-based computation and intelligent graph analytics. Specifically, I firstly propose two theory-based dynamic graph models to characterize temporal trends in large social media networks, then implement and optimize them atop Apache Spark GraphX to improve their performances. In addition, I investigate a semantics-aware optimization framework consisting of offline static analysis and online dynamic analysis on a property graph representing the skeleton of a data-intensive application, to interactively and semi-automatically assist programmers to scrutinize the performance problems camouflaged in the source code. In the design of intelligent graph-based algorithms, I innovate novel neural graph-based approaches with multi-task learning techniques to repair a broad range of programming bugs automatically, and also improve the accuracy of pedestrian navigation systems in only consideration of sensor data of Inertial Measurement Units (IMU, ie accelerometer, gyroscope, and magnetometer). In this dissertation, I elaborate on the definitions of these research problems and leverage the knowledge of graph computation, program analysis, and deep learning techniques to seek solutions to them, followed by comprehensive comparisons with the state-of-the-art baselines and discussions on future research.

Notes

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

Graduation Date

2022

Semester

Summer

Advisor

Wang, Liqiang

Degree

Doctor of Philosophy (Ph.D.)

College

College of Engineering and Computer Science

Department

Computer Science

Degree Program

Computer Science

Identifier

CFE0009244; DP0026848

URL

https://purls.library.ucf.edu/go/DP0026848

Language

English

Release Date

August 2022

Length of Campus-only Access

None

Access Status

Doctoral Dissertation (Open Access)

Download

Share

COinS