Doboli, A.
State University of New York at Stony Brook, NY, USA
Studies in psychology explain that concept combinations are a main step in creative problem solving, including design innovation. This insight is rarely used by modern Computer-Aided Design (CAD) tools to automate design and increase productivity and quality.
This presentation describes a reasoning-based, computational model to produce novel solutions by combining symbolic features, similar to how humans reason in design problem solving. A central component of the model is an associative concept structure that represents the features of existing solutions (the current domain knowledge), and groups the features into concepts with distinct identities and purposes. Different levels of abstraction emerge as a byproduct of grouping. The computational model creates both punctual answers to a problem and more abstract, conceptual solutions.
The associative concept structure of the model is based on work in cognitive psychology and design creativity, including cognitive models for concept feature definition and combination. Our work extends these results into a novel computational approach that tackles a concrete application (electronic circuit design) and can be implemented as a software system.
The proposed associative concept structure defines the similarities and differences between circuit features at the levels of topology, behavior, and performance. Common and distinguishing features are found through matching of the symbolic expressions defining the features. New concepts are added to the structure by combining the features of existing concepts or by exploring new relations between variables. The combined features are associated in the structure to indicate the kind of improvement that they bring to the solution. Concepts are characterized by metrics like variety, utility, novelty, complexity, flexibility, constraining factor, and bottleneck.
A case study shows the systematic creation of new circuit solutions through reasoning based on the associative concept structure, and explains the conceptual advantages of the proposed model as compared to existing CAD tools.