Reimplementation of AMICA in Julia

Abstract

Adaptive Mixture Independent Component Analysis (AMICA) is one of the best performing ICA algorithms. But despite being commonly used, the already existing implementations lack the readability and extensibility needed to allow for future development of the algorithm. This work uses a MATLAB implementation of the AMICA algorithm as reference to develop an easier to understand version. This new implementation is written in the scientific programming language Julia. An evaluation of the correctness, the performance and of how well this implementation achieves it’s goals is provided in this thesis.