You train a deep convolutional neural network for image recognition. After 1 million iterations, the training accuracy is 100% and the validation accuracy does not vary any longer. The algorithm has reached the global optimal. True False You train a deep convolutional neural network using stochastic gradient descent. At the 200th iteration, you evaluated the full gradient and found it exactly zero. At the 200th iteration, the algorithm reached a local minimum. True O False