Evolutionary approach for neural network based agents applied on time series data in the Cloud

Abstract

One of the alternatives and most popular heuristics for global optimisation and for training artificial neural networks is the application of evolutionary searching method. This paper investigates the internal details of the evolutionary searching mechanism from theoretical aspects and also from experimental parallel execution aspects utilising commercial (AWS) and community (ELKH) cloud resources. The paper introduces and evaluates a procedure where the resources assigned for the execution of calculations are scalable horizontally in order to increase significantly the speed of processing.