Genetic algorithms using Galib
Bradley John Hendricks · Digital Scholarship - UNLV (University of Nevada Reno) · 2019
GAlib is a C++ library of genetic algorithm objects that was recently developed at the Massachusetts Institute of Technology. This thesis is to demonstrate its functionality and versatility for implementing haploid tripartite genetic algorithms; We first built a test bed in which GAlib could be used. To achieve this, we used GAlib to solve the Traveling Salesman Problem and implemented two-opt and simulated annealing for compariSon We then examined the use of genetic algorithms for finding loop invariants. We used GAlib successfully to build a model but results remain inconclusive; In our main thrust we applied genetic algorithms to train and develop neural networks. To develop neural network architectures we used two different methods of representing neural networks: connection matrices and graph-generation grammars. We were able to demonstrate that genetic algorithms are an effective tool for training networks as well as for finding network architectures.