Χρήση εξελικτικών αλγορίθμων για την εκπαίδευση Τεχνητών Νευρωνικών Δικτύων
Evolutionary algorithms for training Neural Networks
The current thesis examines evolutionary techniques for the training of Artificial Neural Networks (ANN). The term evolutionary technique refers to a subset of post-heuristic techniques. Evolutionary algorithms include techniques which try to simulate behaviors of populations found in nature, such as the process of evolution of species , the movement of birds from one place to another , finding food in fish-swarm , and more . The conventional way of training an ANN is using the back-propagation algorithm ( BP - Back Propagation). In this work the algorithm BP is compared with 3 evolutionary techniques the Genetic Algorithm ( GA ), the PSO algorithm and a hybrid approach them called HGAPSO . For efficient comparison of these four algorithms we implemented a software which applies the four techniques on data sets . The data sets that were used were recovered from the UCI Machine Learning Repository ( https://archive.ics.uci.edu/ml/datasets.html). The training data properly transformed, in an equivalent form so as to be used by the software.