I think now it became more clear. Thank you again for patience.
You are welcome 
Now that I think you might actually want to discuss it in a neural network setting, for your following question:
It is true that you want more neurons/layers in order to model a higher degree polynomial term.
Cheers,
Raymond
1 Like
Try this Tensorflow playground.
Follow me through the following three examples:
Example 1:
If you choose the first dataset (which is a binary dataset) in the left (circle number 1), pick the degree-2 polynomial features (circle number 2), then you don’t need any hidden layer (circle number 3) and you will still be able to fit it pretty well (circle number 4)
Example 2:
However, if you change to use the degree-1 features (circle number 5), the result will be bad (circle number 6)
Example 3:
If you want the model to learn non-linear features from degree-1 features, what do you do? Add hidden layers and neurons like below. Note that the result still doesn’t look very nice, but I will leave it to you to experiment how to add layers/neurons to get a good result. There are more than one way to do it. It is a pretty interesting exercise actually. Try it!
Good luck!
Raymond
Thank you very much 