H. Muñoz-Avila and F. Ricci (Eds.): ICCBR 2005, LNCS 3620, pp. 163 – 176, 2005. © Springer-Verlag Berlin Heidelberg 2005 Transfer in Visual Case-Based Problem Solving Jim Davies 1 , Ashok K. Goel 2 , and Nancy J. Nersessian 2 1 School of Computing, Queen’s University, Kingston, Ontario, K7L 3N6, Canada jim@jimdavies.org 2 College of Computing, Georgia Institute of Technology, Atlanta, Georgia 30332, USA {nancyn, goel}@cc.gatech.edu Abstract. We present a computational model of case-based visual problem solving. The Galatea model and the two experimental participants modeled in it show that 1) visual knowledge is sufficient for transfer of some problem- solving procedures, 2) visual knowledge facilitates transfer even when non-visual knowledge might be available, and 3) the successful transfer of strongly-ordered procedures in which new objects are created requires the reasoner to generate intermediate knowledge states and mappings between the intermediate knowledge states of the source and target cases. We describe Gala- tea, the two models created with it, and related work. 1 Introduction Experimental evidence shows that visual knowledge often plays a role in case-based reasoning [2,7,11]. Why might this be? What functions do the visual representations serve in retrieval, adaptation, evaluation and storage of cases? These questions are very broad because they pertain to a variety of cognitive phenomena ranging from visual perception to external memory to mental imagery. In order to explore these issues deeply, in the following discussion we focus exclusively on case-based prob- lem solving. Problem solving involves generating a procedure which may contain a number of steps. We will call procedures with the following two properties “strongly- ordered procedures:” 1) two or more steps are involved, and 2) some steps cannot be executed before some other steps have already been executed. Case-based problem solving is taking a solution from a source case and applying that solution or a modifi- cation of it to a target case. Many past case-based systems in problem-solving domains have used visual knowledge and have supported visual reasoning (e.g., ARCHIE [13]. AskJef, [1]). However, these systems typically contain multi-modal cases, i.e., cases that contain both visual (e.g., photographs, drawings, diagrams, animations and videos) and non- visual knowledge (e.g., goals, constraints, plans and lessons). As a result, the precise role of visual knowledge in case-based problem solving remains unclear. In contrast, the present work deals with cases that contain only visual knowledge. Further, past case-based systems such as ARCHIE and AskJef leave the adaptation task to the user and do not automate the transfer of diagrammatic knowledge from a source case to a target problem. The present work directly addresses the transfer task in case-based problem solving.