Robustness of attractor networks and an improved convex corner detector
P. Nachbar, A.J. Schuler, T. Fussl, Josef A. Nossek, Leon Ong Chua · 2003
The authors point out that by defining several notions of robustness for an attractor network, it is possible to augment previous results about the AdaTron algorithm by explicit values for the robustness of the optimal weights. It is shown that the symmetry of a problem is reflected by the invariance of the optimal weights. This enables one to deduce that a convex corner detection, using a discrete-time cellular neural network (DTCNN), cannot be accomplished with just one clock cycle, and an improved convex corner detector is proposed.>