Neural Network Approach to Control System Identification with Variable Activation Functions

Abstract

Human beings epitomize the concept of "intelligent control." Despite its apparent computational advantage over humans, no machine or computer has come close to achieving the level of sensor-based control which humans are capable of. Thus, there is a clear need to develop computational methods which can abstract human decision-making processes based on sensory feedback. Neural networks offer one such method with their ability to map complex nonlinear functions. In this paper, we examine the potential of an efficient neural network learning architecture to the problems of system identification and control. The cascade two learning architecture dynamically adjusts the size of the network as part of the learning process. As such, it allows different units to have different activation functions, resulting in faster learning, smoother approximations, and fewer required hidden units. We use the methods discussed here towards identifying human control strategy.