Skip to Main Content
Almost all of the learning paradigms used in machine learning, learning automata (LA), and learning theory, in general, use the philosophy of a student (learning mechanism) attempting to learn from a teacher. This paradigm has been generalized in a myriad of ways, including the scenario when there are multiple teachers or a hierarchy of mechanisms that collectively achieve the learning. In this paper, we consider a departure from this paradigm by allowing the student to be a member of a classroom of students, where, for the most part, we permit each member of the classroom not only to learn from the teacher(s) but also to ldquoextractrdquo information from any of his fellow students. This paper deals with issues concerning the modeling, decision-making process, and testing of such a scenario within the LA context. The main result that we show is that a weak learner can actually benefit from this capability of utilizing the information that he gets from a superior colleague-if this information transfer is done appropriately. As far as we know, the whole concept of Students learning from both a teacher and from a classroom of Students is novel and unreported in the literature. The proposed student-classroom interaction has been tested for numerous strategies and for different environments, including the established benchmarks, and the results show that students can improve their learning by interacting with each other. For example, for some interaction strategies, a weak student can improve his learning by up to 73% when interacting with a classroom of students, which includes students of various capabilities. In these interactions, the student does not have a priori knowledge of the identity or characteristics of the students who offer their assistance.