Minimax Forward and Backward Learning of Evolving Tasks with Performance Guarantees Santiago Mazuelas