Learning in Non-stationary Conditions: A Control Theoretic Approach