Title:  Quantitative Analysis of Adaptiveness and Consistency of a Class of Online Learning Algorithms

Committee: 

Dr. Fumin Zhang, ECE, Advisor

Dr.

Dr.

Abstract:

Expert based learning algorithms have been used by robots to choose satisfying reactions to human movements. For simple algorithms with two outputs and two experts, Markov chains have been used to rigorously quantify their adaptiveness and consistency. This paper proposes to expand the complexity of algorithms that can be analyzed for consistency and adaptiveness. This includes adding more experts and more outputs to allow for more potential robot responses, and utilizing the learned human response to identify the internal states of a human. This paper also proposes considering the human's response to learning so that the resulting robot actions do not have chatter. Finally this paper proposes physical experiments to show the learning algorithms in action.