Stochastic Learning and Optimization: A Sensitivity-Based