Any chance the guy who did this work is called Alex?

@shivalgo Yeah it was named after Alex Krizhivsky. See: https://en.wikipedia.org/wiki/AlexNet.

What is the point of having multiple conv/reLU layers? Don't do they all do the same thing?

@hamood, having more conv/reLU layers means having more weights, so the model can represent more complicated functions. This is helpful if you want your model to do something pretty complex, e.g. classify lots of images. However, adding more layers can lead to overfitting, where the model starts to just memorize the training data, and as said elsewhere in the lecture, the architecture of the model also matters.

What are some other ways to introduce non-linearity?