Can anyone explain me why in the bottom it’s written “dont use linear activations in hidden layers” … I mean if we use linear activations in hidden layers and sigmoid in output…we anyways get a logistic regression model…so what’s the issue?
Hi @gigaGPT,
That’s the issue. We don’t want the neural network to end up behaving like a logistic model, where it simply computes a linear combination of ‘z’ values (e.g in hidden layers) and applies a sigmoid transformation. In such cases, using a neural network wouldn’t make much sense. That’s why we incorporate non-linear activations in the hidden layers of neural network, as they enable exploration of non-linearity.
3 Likes
Ohkk…thanks 
, I got it