Cooperative System Identification via Correctional Learning

Abstract We consider a cooperative system identification scenario in which an expert agent (teacher) knows correct, or at least good, model of the and aims to assist learner-agent (student), but cannot directly transfer its knowledge student. For example, teacher’s might be abstract teacher student employing different classes, renders parameters uninformative In this paper, we propose correctional learning as approach above problem: suppose that order student, can intercept observations collected from modify them maximize amount information receives about system. formulate general solution optimization problem, for multinomial instantiates itself integer program. Furthermore, obtain finite-sample results on improvement assistance (as measured by reduction variance estimator) binomial numerical experiments, illustrate proposed algorithms verify theoretical have been derived paper.