Um estudo comparativo de algoritmos de ICA em alfabetos finitos sob modelos lineares e não-lineares

Recently proposed ICA over finite alphabets algorithms are analysed in this work, whether they are under the linear paradigm i.e. finite fields, whether they comprise a permutation based model, namely the QICA algorithm. Numerical experiments are performed for the classical application of BSS, and for a non-linearly-generated dataset. Despite the QICA method can potentially provide more robust solutions, the results indicated a superior performance on few scenarios, with a higher average execution time. The BSS experiment confirmed AMERICA algorithm as the most efficient in terms of separation, with the lowest absolute values of execution time.