Master of Science in Computer Engineering, University of Dayton, 2017, Electrical and Computer Engineering

This thesis proposes the use of a genetic algorithm (GA) to optimize the accuracy of a convolutional neural network (CNN). The GA modifies the structure of the CNN such as the number of convolutional filters, strides, kernel size, nodes, learning parameters, etc. Each modification of the network is trained and evaluated. Mutation of evolved networks create more successful networks over multiple generations. The final evolved network is 4.77% more accurate than a network pro- posed in the previous literature. Additionally, the evolved network is 13.4% less computationally complex.

Committee: Eric Balster (Advisor); Tarek Taha (Committee Member); Frank Scarpino (Committee Member) Subjects: Artificial Intelligence; Computer Engineering; Computer Science

Master of Science, The Ohio State University, 2011, Electrical and Computer Engineering

Dielectric rod antenna has been shown to provide wideband, dual-polarization, and symmetric pattern. A two layer dielectric rod antenna (DRA) was shown to achieve more than 4:1 bandwidth of stable gain and with more than 55 degree HPBW. Although, it was predicted that a multi-layer DRA design should be able to achieve more than 8:1 bandwidth, such design has never been realized due to its design and fabrication complexity. This thesis discusses about how we overcome these challenges via the utilization of genetic-algorithm (GA) for design optimization procedure. The key challenge in designing a multi-layer DRA is to choose proper thickness and dielectric constant of each layer to meet desired VSWR, pattern, and phase center requirements. This becomes even more difficult when the when the number of layers increases for achieving a greater bandwidth. Although most of the new commercial EM simulation codes such as FEKO and HFSS have built-in parameter optimization functions, this approach consumes too much computer resources and is not easy to implement complex design goal that involves many parameters and different frequencies. This thesis will demonstrate an alternative antenna design optimization method that employs an external GA program which interacts with commercial EM simulation software such as HFSS, to automatically choose optimized thicknesses and dielectric constants of multi-layer DRA that achieves constant gain, beamwidth, phase center and wideband performance objectives.

Committee: CHI-CHIH CHEN (Advisor); JOHN VOLAKIS (Committee Member) Subjects: Electromagnetics

Master of Science (MS), Ohio University, 2008, Industrial and Systems Engineering (Engineering and Technology)

In this thesis, a new Vehicle Routing Problem with time windows is introduced where driving time and working time restrictions are imposed on the problem. The objective is to minimize the total traveling distance. A mathematical model and also a genetic algorithm-based solution technique are proposed first and later they are compared. The results show that GA can find the optimal solution with a high frequency for small problems. Two new genetic operators (route-sensitive crossover and route-sensitive mutation) are also proposed to solve this particular problem. The results of the experiments also show that the proposed crossover operator and swap mutation outperformed the classical operators.

Committee: Gursel A. Suer (Advisor) Subjects: Engineering; Industrial Engineering