A Kohonen Map for Temporal Sequences

This paper presents a new kind of Kohonen self-organizing maps, designed to deal with temporal sequences. The approach relies on spaces of functions: instead of the usual spatial inputs, the map takes its inputs in the space of functions of time having their values in a given vectorial space. This space of functions is clustered by the map which, doing so, takes into account what precisely happened during time instead of only using the final spatial result. After a presentation of the theoretical basis of the approach, its application to signature verification is presented. It consists of classifying a given number of signatures using their spatial and temporal characteristics. The classification is then used to authenticate or reject new incoming signatures.