Optimal functional supervised classification with separation condition (Q2174981)

The paper deals with binary supervised classification of trajectories of stochastic processes. Basing on the finite training set of trajectories one has to determine to which of two unknown processes the observed trajectory is related. In this paper the authors consider the trajectories of solutions of stochastic differential equations \[ dX_t = \mu(t) dt +dW_t, \qquad t \in [0,1],\quad \mu \in \{f,g\}, \] where the drifts \(f\) and \(g\) belong to the Sobolev-Hilbert space \(H_s (0,1)\), \(s\geq 0\), and are significantly different \(\|f-g\| \geq \Delta >0\). The goal is to provide a natural classifier such that asymptotically no other classifier is significantly outperforming it.