Visual signal representation for fast and robust object recognition

Internal signal representation is critical for accurate perception, memory, and rapid recall in natural intelligence. In this paper we consider a model for cognitive computation that has as its base an algorithm for converting spatial signals into temporal representations. The model is motivated by vision functions but is supported by applications to concept abstraction and analogical thinking, as well as experimental comparison with the pixel intensity representation of visual images. In fact our method has demonstrated ability to defend against a well-known adversarial attack on the MNIST digits recognition. Ordinal optimization is important in achieving quickness of the algorithm.