Applications of Quantum Probability Theory to Dynamic Decision Making.

摘要

The broad term goal of this research program was to build a foundation for constructing probabilistic-dynamic systems from principles based on quantum as opposed to classical probability theory. So far we have applied these principles to both traditional, one-stage decision problems studied by decision researchers as well as dynamic Markov decision problems used in computer science and engineering. The more specific goal of the proposed research was to develop new applications of quantum probability applied to dynamic decision situations: (a) To develop a quantum reinforcement learning model for learning a sequence of actions in a Markov decision problem environments that is fast learning and robust with respect to changes in the environment; (b) To theoretically derive the convergence and speed of convergence properties of the new quantum learning algorithm for the dynamic environments; and most importantly, (c) To experimentally test whether the quantum reinforcement learning model provides a better account of actual human performance in Markov de