New BFA method based on attractor neural network and likelihood maximization

Abstract

What is suggested is a new approach to Boolean factor analysis, which is an extension of the previously proposed Boolean factor analysis method: Hopfield-like attractor neural network with increasing activity. We increased its applicability and robustness when complementing this method by a maximization of the learning set likelihood function defined according to the Noisy-OR generative model. We demonstrated the efficiency of the new method using the data set generated according to the model. Successful application of the method to the real data is shown when analyzing the data from the Kyoto Encyclopedia of Genes and Genomes database which contains full genome sequencing for 1368 organisms.