Evolving Artificial Neural Networks that Develop in Time

Although recently there has been an increasing interest in studing genetically-based development using Artificial Life models, the mapping of the genetic information into the phenotype is usually as an abstract process that takes place instantaneously, i.e. before the creature starts to interact with the external world and is tested for fitness. In this paper we show that the temporal dimension of development has important consequences. By analyzing the results of simulations with temporally developing neural netwoks we found that evolution, by favouring the reproduction of Os which are efficient at all epochs of their life, selects genotypes which dictate early maturation of functional neural structure but not of nonfunctional structure. In addition, we found that development in time forces evolution to be conservative with characters that mature in the first phases of development while it allows evolution to play more freely with characters that mature later in development. Finally,...

• Type: Conference
• Year: 1995
• Where: ECAL
• Authors: Stefano Nolfi, Domenico Parisi
• Keywords: Artificial Intelligence, ECAL 1995, Functional Neural Structure, Genetically-based Development, Time Forces Evolution