UNIVERSITY OF HERTFORDSHIRE
     COMPUTER SCIENCE RESEARCH COLLOQUIUM

                presents

  "Biologically and Cognitively inspired
           General Game Playing:
    Self-motivation from hidden structure in
     Sokoban, Gambler's Problem and Pacman"

       	      Tom Anthony
     (Adaptive Systems Research Group,
   School of Computer Science, University
	         of Hertfordshire)

     	10 November 2010 (Wednesday)
         Lecture Theatre A154
	  Hatfield, College Lane Campus
	         	1 -2 pm


      Everyone is Welcome to Attend

      Refreshments will be available



Abstract:

  We use empowerment, a recently introduced information theoretic
biologically inspired method for optimizing the flow of information
through an agent's perception-action loop, allowing an agent with an
unknown task to assign expected value to states in the environment. We
demonstrate how an extension to this model allows an embodied agent to
detect the dynamics and relevant structures in its environment that can
be used, in conjunction with a method for clustering action sequences
into `strategies', to maintain this quantity by selecting `preferred'
action sequences. We show how this method provides a proto-heuristic
for non-terminal states and can be used for self-motivation in taskless
scenarios. We demonstrate, using a Sokoban-inspired box-pushing scenario,
the Gambler's Problem, and a simplified Pacman game how this technique
may prove to be a promising pathway to improved AI game playing.

---------------------------------------------------
Hertfordshire Computer Science Research Colloquium
http://homepages.stca.herts.ac.uk/~nehaniv/colloq