In 1994, Robert Goldstone described this model in his article, "Similarity, Interactive Activation, and Mapping" in the Journal of Experimental Psychology. SIAM used a connectionist architecture to model human judgments of the similarity between two scenes, showing the sensitivity to structure which previous models ignored. SIAM uses a connectionist architecture which doesn't learn, but where constraints spread through the network exciting consistent correspondences and inhibiting inconsistent ones.
This tool provides a web-based interface to a simplified version of the model. (It was based on the excerpt in Polk and Seifert's "Cognitive Modeling" book, and ignores spatial relations.) Users may specify scenes using Goldstone's butterfly stimuli, and run the network for a set number of cycles. It is also possible to edit the representation to add or subtract attributes and values. I meant to include a variant where I use the model along with LSA to measure the similarity between sentences, but haven't gotten there yet.
Go to the demoemail me