A common framework for distributed representation schemes for compositional structure

Over the last few years a number of schemes for encoding compositional structure in distributed representations have been proposed, e.g., Smolensky's tensor products, Pollack's RAAMs, Plate's HRRs, Halford et al's STAR model, and Kanerva's binary spatter codes. All of these schemes can placed in a general framework involving superposition and binding of patterns. Viewed in this way, it is often simple to decide whether what can be achieved within one scheme will be able to be achieved in another. Furthermore, placing these schemes in a general framework reveals unexplored regions in which other related representation schemes with interesting properties.