Value learning-based algorithms