MS, University of Cincinnati, 2023, Engineering and Applied Science: Computer Science

Generative Adversarial Networks (GANs) are canonically characterized as a generative model for unsupervised learning where two neural networks compete against each other with the loss of one agent being the gain of the other. Undesired artifact formation is a common problem that occurs during the training process of high-resolution, image generating GANs. These artifacts can appear in several forms, such as checkerboard-like patterns or blurry features that disrupt training and prevent optimal image generation. This work will look at how these artifacts arise and explore several strategies that attempt to mitigate their formation during the training of a DCGAN.

Committee: Badri Vellambi Ravisankar Ph.D. (Committee Chair); Raj Bhatnagar Ph.D. (Committee Member); Jillian Aurisano Ph.D. (Committee Member) Subjects: Computer Science

Master of Science in Engineering (MSEgr), Wright State University, 2014, Biomedical Engineering

Introduction: Functional near-infrared spectroscopy (fNIRS) is a relatively young technique in the field of medical imaging. As such, it has yet to be widely implemented for clinical use, despite its promising advantages. However, unlike fMRI-its much bulkier and costly counterpart-fNIRS has yet to be proven as a standalone imaging tool within a clinical setting, particularly that of ophthalmology or physical therapy. Methods: Ten healthy young adults (23.8 ± 4.8 years) participated in the study. Activation of the visual cortex was achieved utilizing various reversing checkerboard stimuli across three data collection sessions for each participant. Further, activation of the motor cortex was achieved using simple grasping and finger tapping tasks. Data was processed with MATLAB scripts and statistical analysis was performed using JMP. Results: Quantitatively, statistically significant differences in the level of activation were elicited by some stimuli, but not others. No differences were discovered between the levels of activation for the two motor tasks. However, as expected, differences were observed between the hair types of participants for both visual and motor activation. Additionally, one of the three data collection sessions for each participant tended to give statistically different results than the other two. Qualitatively, the number of stimulus events and data channels which showed activation were inconsistent. Conclusions: It has been shown, both previously (by others) and within this study, that fNIRS is indeed feasible for investigating the visual and motor cortices. However, a reliable level of robustness and sensitivity is required for clinical implementation. This research shows that fNIRS can in fact achieve an appropriate level of sensitivity for visual studies, but it still lacks an appropriate level of robustness in terms of repeatability and corporal differences for assessment of visual or motor dysfunction.

Committee: Nasser Kashou Ph.D. (Advisor); Frank Ciarallo Ph.D. (Committee Member); Thomas Hangartner Ph.D., FAAPM (Committee Member); Ping He Ph.D. (Committee Member) Subjects: Biomedical Engineering; Biomedical Research

Master of Science in Engineering (MSEgr), Wright State University, 2009, Electrical Engineering

The objective of this thesis is to demonstrate the layered sensing concept using Master-Slave cameras. The process of 2D camera calibration and several key factors that can present error during such calibration are described. The analysis and results are based on calibration of a pinhole model camera system. The calibration is accomplished using OpenCV software and the results are analyzed using MATLAB software. These results are divided into intrinsic and extrinsic camera parameters. These parameters are then used to determine the position and orientation of the object in the camera coordinate system. This thesis also explores the use of two cameras as a Master-Slave system to demonstrate the layered sensing concept. The Master camera's orientation, zoom, and distance from the Slave camera is fixed. Using the position and orientation of the object in the Master camera coordinate system, the position of the object in the Slave coordinate system is found using transformation matrices. This information is then used to determine the pan/tilt angles of the Slave camera. The Slave camera is then directed to focus on the object using servo control.

Committee: Kuldip S. Rattan PhD (Advisor); Devert Wicker PhD (Committee Member); Doug Petkie PhD (Committee Member) Subjects: Electrical Engineering

PhD, University of Cincinnati, 2008, Arts and Sciences : Physics

We present results of three studies done using a dynamical cluster quantum Monte Carlo approximation. First, we investigate the d-wave superconducting transition temperature Tc in the doped 2D repulsive Hubbard model with a weak inhomogeneity in hopping in the form of checkerboard pattern or a lattice of 2×2 plaquettes. Near neighbor hoppings within a plaquette is t and that of between the plaquettes is t'. We investigate Tc in the weak inhomogeneous limit 0.8t < t'< 1.2t. We find inhomogeneity (t'= t) suppresses Tc. The characteristic spin excitation energy (effective exchange energy) and the strength of d-wave pairing interaction decrease with decreasing Tc. The latter observations suggest a strong correlation among effective exchange interaction, Tc and the d-wave pairing interaction of the system. Second*, we further find that enhancement of effective exchange interaction causes a slight increase in Tc of a weakly disordered system with low impurity concentration, compared to the homogeneous system. Here the disorder is introduced to homogeneous repulsive 2D Hubbard model as a weak local potential disorder. Third, we present an improved maximum entropy method to analytically continue quantum Monte Carlo data with a severe sign problem. * A result from a collaborative study done with A. Kemper of Florida State University.

Committee: Mark Jarrell (Committee Chair); F. Paul Esposito (Committee Member); Michael Ma (Committee Member); Leigh M. Smith (Committee Member); Robert J. Endorf (Committee Member) Subjects: Physics

MS, University of Cincinnati, 2003, Engineering : Mechanical Engineering

Topology optimization is fast emerging as an integral part of the product development cycle using Computer Aided Engineering tools. The optimal structure and shape of a product can be predicted in the initial stages of a development cycle using topology optimization. The goal of topology optimization is to find the best distribution of material for a structure such that an objective criterion, like global stiffness, takes on an extremum value subject to given constraints. These constraints are typically placed on the volume. In this thesis, some of the numerical issues that occur in the solution of a topology optimization problem are discussed. These numerical issues include the formation of checkerboard patterns in the final topology and sensitivity of the optimal solution to the mesh size used to discretize a domain. A computationally cheaper heuristic filtering scheme to counter these numerical instabilities is studied. The effects of non-conforming or discontinuous Galerkin finite element formulations to solve problems in topology optimization are also studied. Several numerical experiments involving the use of bilinear and biquadratic finite elements for the solution of the topology optimization problem are presented. In addition, an application area referred to as the “inverse homogenization procedure” using the topology optimization procedure for the design of materials with prescribed material properties is examined.

Committee: Dr. Kumar Vemaganti (Advisor) Subjects: