To machine learning pioneer Terry Sejnowski, the mathematical technique called stochastic gradient descent is the “secret sauce” of deep learning, and most people don’t actually grasp its ...