To give an example. Let's just consider the MNIST dataset of handwritten digits. Here are some things which might have an impact on the optimum model capacity: There are 10 output classes The inputs are 28x28 grayscale pixels (I think this indirectly affects the model capacity. eg: if the input...