Unfalsified control: a behavioral approach to learning and adaptation

Unfalsified control theory facilitates the representation of adaptive processes of control law discovery from evolving information flows and noisy data. In the paper, the theory of unfalsified adaptive control is examined from the behavioral perspective of Willems (1991). An abstract, but parsimonious, min-max optimization problem formulation is developed that describes and unifies direct adaptive control, learning theory and system identification problems in a common behavioral setting based on the concept of controller/model unfalsification. Thus, adaptive control is seen to be firmly and directly linked to, and to conceptually unified with, the growing body of knowledge on behavioral approaches to model validation and unfalsified system identification. The results elucidate and underscore the fertile conceptual links that exist between adaptive control theory and the rich theory of system identification