Unconventional Methods in Big Data Analysis

Abstract

Technological progress and growing computing power are causing data avalanche in almost all sciences, including astrophysics. The general goal of the data classification is to find patterns in the data and translate these patterns into useful information. This work deals with classification of Be stars data, showing number of different shapes of the emission lines that reflect underlying physical properties of a star. The speed of the current classification methods is not acceptable to classify a huge amount of the data and therefore this work proposes an innovative classification method based on evolutionary algorithms, symbolic regression, irregular dynamics, and massively parallel computation. Second part of this work is focused on the data classification using artificial neural networks synthesized by means of symbolic regression, evolutionary algorithms, or network growth model. The synthesized networks' behavior dynamics is investigated and analyzed, and idea about similarity between swarm and neural based algorithms is proposed.