Learning Skill and Representation Hierarchies for Eective Control Knowledge Transfer

Learning capabilities of computer systems still lag far behind biological systems. One of the reasons can be seen in the inecient re-use of control knowledge acquired over the lifetime of the articial learning system. To address this deciency, this paper presents a learning architecture which transfers control knowledge in the form of behavioral skills and corresponding representation concepts from one task to subsequent learning tasks. The presented system uses this knowledge to con-struct a more compact state space represen-tation for learning while assuring bounded optimality of the learned task policy by uti-lizing a representation hierarchy. To demonstrate this control knowledge trans-fer, a sequence of experiments in a video game domain is presented, illustrating its ability to reduce representational complexity and to enhance learning performance. 1