In Machine Learning (ML) course, classifier hypothesis is: h = theta’ × X
In week 2 of Neural Network and Deep Learning, the 1st part of DLS: h = omega’ × X + b
What is purpose of b parameter?
1 Like
Hypothesis will contain b i.e. the bias term.
Please read this keeping in mind that y_0 is b
1 Like
Yes, Prof Ng has chosen to use a different notation in the DLS courses than he used in the Stanford ML course. The ML course was published in 2011 and DLS in 2017. The whole field of ML/DL evolved a lot in those years.
In ML, he makes the bias units implicit and \theta_0 is the bias coefficient. In DLS, he has chosen to make the bias units explicit and separately notated. I think the DLS notation ends up making more sense. When you get to real Neural Networks in Week 3 and Week 4 of ML, remember all the confusion about whether you need to add the ones for the bias units or whether they are already there and then having to remove them at some points during back prop. Here in DLS with the bias units explicitly separate everything is just simpler to understand.
Of course you can have your own opinion about which is better, but Prof Ng is the boss. We just have to understand his notation and “deal with it”.
2 Likes
Thank you @paulinpaloalto for your explanation. so b in DLS is the bias which was theta0 in ML. The bias is used explicitly to remove all the confusion it had. If so, that means there is no performance improvement by using b, the purpose here is merely making it to be more readable and convenient, right?
Correct! There is no difference at all from a mathematical point of view. It’s just a notational change. The results will be the same, just “packaged” in a slightly different form.