Approximating general Markovian decision-problems by clustering their state- and action-spaces
Willibald Doeringer · Mathematische Operationsforschung und Statistik Series Optimization · 1984
Following ideas of V. Nollau and A. Hahnewald-Busch (cp. [17], [18], [19]) we develop a method of clustering state- and action-spaces of General Markovian Decision Problems into finitely many subsets serving as states and actions of approximating finite Markovian Decision-Problems. And we show that under suitable assumptions the value-functions of our constructed problems approximate those of the original one in a uniform sense.