← Learning Logistic Regressors by Gradient Descent