Artificial Neural Networks, Symmetries and Differential Evolution

摘要: Neuroevolution is an active and growing research field, especially in times of increasingly parallel computing architectures. Learning methods for Artificial Neural Networks (ANN) can be divided into two groups. mainly based on Monte-Carlo techniques belongs to the group global search methods, whereas other such as backpropagation belong local methods. ANN's comprise important symmetry properties, which influence On hand, are generally unaffected by these symmetries. In literature, dealing with symmetries reported being not effective or even yielding inferior results. this paper, we introduce so called Minimum Global Optimum Proximity principle derived from theoretical considerations breaking, applied offline supervised learning. Using Differential Evolution (DE), a popular robust evolutionary optimization method, experimentally show significant efficiency improvements breaking.