A learning framework where two processes update at different speeds to ensure stable convergence of interdependent estimates.