Neural networks (NN) have become a central component in many, if not most of, machine learning and Artificial Intelligence systems. However, studies have shown that these models are easily fooled and not robust against adversarial attacks. In this dissertation, we study the robustness of such models by exploring three directions. Then, we extend the synthesis problem of NNs to propose a solver for some combinatorial optimization problems. The first direction, we investigate adversarial attacks on two hierarchical classification (HC) models: the Flat HC (FHC), and the Top-Down HC (TDHC). For the second direction, using the concept of grouping of labels and coarse labels, we formalize a new notion of coarse robustness that is defined with respect to a specified grouping of the class labels.
The third, more general, direction is the Bidirectional One-Shot Synthesis (BOSS) problem for adversarial examples. Subsequently, we extend the synthesis problem of adversarial attacks to solving combinatorial optimization problems. This is accomplished by presenting NN structures derived with respect to finding Maximal Independent Sets in the graph of interest. The success of machine learning solutions for reasoning about discrete structures has brought attention to its adoption within combinatorial optimization algorithms. Such approaches generally rely on supervised learning by leveraging datasets of the combinatorial structures of interest drawn from some distribution of problem instances.
Major: Electrical Engineering
Educational Career:
Bachelor's of Electronics and Communications Engineering , BS, 2009, University of Baghdad
Master's of Electrical and Computer Engineering , MS, 2013, San Diego State University
Committee in Charge:
Wasfy Mikhael , Chair, Electrical and Computer Engineering
George Atia, Co-Chair, Electrical and Computer Engineering
Issa Baterseh, Electrical and Computer Engineering - UCF
Marianna Pensky , Department of Mathematics - UCF
Yaser Pourmohammadi Fallah, Electrical and Computer Engineering - UCF
Approved for distribution by Wasfy Mikhael , Committee Chair, on August 26, 2023.
The public is welcome to attend.
Read More