Jack How does Stochastic Gradient Descent differ from traditional batch gradient descent? What benefits can SGD offer when training models on large datasets? How do batch size, learning rate, and noisy updates affect training speed and model performance?