Real-time Topology Learning

The demonstration shows the first implementation (in any technology) of a real-time system for learning the spatial topological arrangements of cellular elements using an STDP-based unsupervised learning rule. Given a stream of address-event labelled inputs from the cellular elements, the algorithm learns their adjacency and automatically detects edge elements within a short time. The demonstration graphically shows the progress of learning during user interaction. Related paper: Boerlin, M and Delbruck, T and Eng, K (2008). {\em Getting to Know Your Neighbors: Unsupervised Learning of Topography from Real-World}, Event-Based Input. Neural Computation.