'Weight initialization in CNN, I can't understand initiating all weights as 0 leads to update all W's with the same amount

let f be a liner function by back propagation, since W's update is determined by '-dL/dW', and 'df/dW' is determined by 'x', (since dL/dW = dL/df * df/dW, and df/dW = x) Isn't it it the X that affects W's updates?

I don't get why setting all W's with 0, leads to updating W's with all the same amount.

conv-neural-networkbackpropagation


Sources

This article follows the attribution requirements of Stack Overflow and is licensed under CC BY-SA 3.0.

Source: Stack Overflow

Solution Source

Related Questions