Transient response functions for graph structure addressable memory

While neurons in the brain are individually quite slow, collectively they can recognize concrete objects as well as abstract concepts very quickly. Motivated by this puzzling fact we propose biologically plausible algorithms that are capable of quickly recognizing similar graph structures. Graphs are combinatorial constructions and pose serious challenges to similarity testing. In this paper we use the transient behavior of random walk over graphs to compare their spectral resolution. We collect data from intrinsically parallel random walks to form a graph response function as an effective measure of graph similarity. Our algorithm could be a solution to the long standing mystery of content addressability in the brain.