Feedback stabilization of control policy selection in data/knowledge based systems
Knowledge and/or data-based decision support systems can be viewed as complex control systems designed to optimize a desired criterion. As such, they aid users in transforming objects with an initial state vector to some desired outcome state by application of a suitable control policy. System effectiveness can be measured by the number of decision errors. Completeness and validity of the knowledge base can be facilitated by coupling a database of instances of system use with the implementation of the knowledge base. Instances of system use are viewed as traces which should be consistent with the knowledge base. The database contains records characterizing each instance of system use (a trace) and includes the state vector of the object being controlled, the selected control strategy, and the resultant outcome state vector. Stability of control policies is improved within the decision process by incorporating a feedback data path that is derived from actual outcomes resulting from application of a control policy.
