Human acts learning through lifeline by the neural networks with the deep self-organization

Yulkova Viktoriya Mihailovna, Shilovskii Georgii Vladimirovich

If to overview autonomic robots and systems, learning during the lifeline is the basis to replenishing and changing knowledge by means of multiple trials and experience at all. But be that as it may, the standard models of neural motion with deep recognition by video do not take into account training throughout the entire life cycle, but they study pre-defined training data packages with a specified number of samples and classes of actions. And this means that there is a need to create such training systems that can process and adapt the perceptual signals available at the time as they arrive. In this scientific article, we offer you a description of neural architecture with self-organization, capable of a gradual study of the classifications of human actions on video materials. The architecture includes growing networks of self-organization that have reverse neurons for processing dynamically unstable patterns. The method involves sets of reverse networks with hierarchical organization to introduce the possibility of studying the essence of actions with continuously increasing spatio-temporal fields of susceptibility without human control. With the help of neural dynamics, based on predicting the growth and adaptability of reverse networks by determining the ability to reconstruct incoming sequences, arranging them over time, learning is gained throughout life.

Keywords:lifeline learning, motion recognition, deep learning unattended, self-organization