Recognition of a stream of random events (Q1067792)

We consider the recognition problem for a stream of directly unobservable events (from a finite number of classes) which occur at random moments on the discrete time axis. It is assumed that each event causes a change in the probability properties of the sequence of the observed values of the attribute vector relative to previous values, and this change is associated with the class of the current event. For the proposed class of loss functions, the recognition problem reduces to an extremal problem of optimal partitioning of the sample fragment in the observation interval.