In this paper we propose a new encoding scheme utilises predictive coding technique in order to increase the efficiency of evolving artificial neural network. The predictor encodes the sample data fed to the system and the artificial neural network acts as the decoder. The latter is trained using a data model created via predictive coding, which is generated from the initial sample. Only the residual data output from the encoder is fed to the artificial neural network for authentication. Distributed and local processing has been simultaneously used in parallel and in synchrony. Comparison of the simulation results with those obtained using traditional methods such as selective biometric features shows an improvement in efficiency of up to 80% while utilising a lower complexity neural network.

No abstract

Abstract. In this paper, we focus on the learning aspect of NEAT and its variants in an attempt to solve benchmark problems through fewer generations. In NEAT, genetic algorithm is the key technique that is used to complexify artificial neural network. Crossover value, being the parameter that dictates the evolution of NEAT is reduced. Reducing crossover rate aids in allowing the algorithm to learn. This is because lesser interchange among genes ensures that patterns of genes carrying valuable information is not split or strayed during mating of two chromosomes. By tweaking the crossover parameter and with some minor modification, it is shown that the performance of NEAT can be improved. This enables NEAT algorithm to evolve slowly and retain information even while undergoing complexification. Thus, the learning process in NEAT is greatly enhanced as compared to evolution

This paper presents promising results achieved by applying a new coding scheme based on predictive coding to neuroevolution. The technique proposed exploits the ability of a bit, which contains sufficient information, to represent its neighboring bits. In this way, a single bit represents not only its own information, but also that of its neighborhood. Moreover, whenever there is a change in bit representation, it is determined by a threshold value that determine the point at which the change in information is significant. The main contributions of this work are the following: (i) the ratio of the number of bits to the amount of information content is reduced; (ii) the complexity of the overall system is reduced as there is lesser amount of bit to process; (iii) Finally, we successfully apply the coding scheme to NEAT, which is used as a biometric classifier for the authentication of keystroke dynamics General TermsArtificial intelligence, Neural Network, Neuroevolution.

No abstract