Exercise 7.1.10

If we play with the example 2 at http://playground.tensorflow.org/, we can see that with 2 hidden layers:

• 8 + 2: is much easier to find the solution, takes much less epochs to separate all the points.
• 2 + 8: usually takes much longer to find a solution, and the solution can’t completely separate all the points.

So we should put more of neurons in layer 1 instead of layer 2.