Deep Convolutional Neural Networks for multimedia applications

Over the last few years, we can observe the extremely rapid development of multimedia understanding techniques. Practically all of them are based on Convolutional Neural Networks (CNNs) that appear to be an obligatory processing block in a modern cognitive architecture. The success of CNNs is not synchronized with development of diagnostic tools and CNNs understanding. Due to many number of layers and number of weights, in most applications, CNNs are treated as a kind of black box algorithm. In this work we present currently used methods in Deep Learning diagnostics, as well we propose own methods of more detailed analysis of neural activity.