For intuition 1, Andrew said that " If
lambda is too large - a lot of weights will be close to zeros which will make the NN simpler (you can think of it as it would behave closer to logistic regression). "
Why when a lot of weights will be close to zeros, the NN will be behave close to logistic regression?
If W (weights) are close to 0 then g(Wx+b) is almost g(b), the sigmoid of a bias term, which as we move deeper to the neural network becomes the computation of sigmoid of a previous number, because the weights are negligible. So basically its a just the sigmoid of number as far as i understand it.
so what Andrew meant is because a lot of neurons will shut down, the NN will be simpler as if it is "a logistic regression NN (because logistic regression NN has one layer )?
That means “you can think of it as it would behave closer to logistic regression” is just to make us visualize how simple the NN will be ?
Yeah i think thats right what you say.