As of 2023-08-09 the slide for the derivates from backpropagation in the video “Training a CBOW Model: Backpropagation and Gradient Descent” seems to be totally different from an earlier version in 2023-02-22.
In the newer version a step-function is showing up in the formulas but is not explained on the audio track. Additionally the last formula for b2 is not shown. If you want to apply batch learning somewhere, you therefore need to guess b2 and do your research about the step function. I am not even confident if the formulas are correct at all.
I have found a screenshot of “jennynz” in the thread Course 2 week 4 assignment exercise 4 incorrect b1 b2 gradients - NLP with Probabilistic Models / Week 4 NLP with Probabilistic Models - DeepLearning.AI.
There are the original derivates shown. Note that they do not have a step function, the b2 formula is listed and the w1 and b1 formulas make use of the ReLu function in contrast to the “updated” slide.
I highly recommend to refer to this older screenshot if you need to apply these functions somewhere and omit the “updated” slide.