Lifelong Learning Repetitive Neuro-Controller

Abstract

Over the past five years, almost every application of engineering has tried to implement an artificial neural network to improve performance in one way or another. The study of real neurons in the human brain on the other hand, which paved the way for such applications, has been ongoing for well over five decades. While significant progress has been made, artificial neurons fall well short of the capabilities of real neurons in some key areas, such as the ability to perform lifelong learning. This limitation can be traced back to fundamental differences in network architecture and overall implementation. To remedy this, attention must first be directed back to the foundational science behind how the human brain acquires, stores, uses, and selectively removes specific memory or knowledge. Next, novel architectural concepts and implementation strategies to address the aforementioned limitations must be developed. Here, this will be done through careful considerations of all aspects within the repetitive regime. The human brain, by design, is very reliant on continual lifelong learning to solve problems. Control strategies used today, artificial neural networks, and even combinations of both are unable to optimally engage in this fundamental process. This paper attempts to bridge the gap and push toward enhanced lifelong learning control strategies for future use.