I’m doing the weight initialization lab where we see that initializing weights to zero causes the network to fail to break symmetry. For context here’s the text I’m looking at: [Capture] I don’t understand the last sentence. Why does the prediction always being 0.5 mean that the weights don’t get…

Why don't weights get adjusted when initialized to 0?

paulinpaloalto August 30, 2023, 1:05am 3

Here’s a thread which discusses Symmetry Breaking in general. Here’s another one.

Topic		Replies	Views
Week 1, Programming Assignment initialization, Exercise 1 - initialize_parameters_zeros Improving Deep Neural Networks: Hyperparameter tun	8	825	December 15, 2023
Zero initialization of weights Improving Deep Neural Networks: Hyperparameter tun	2	668	October 27, 2021
Concept in Initialization Assignment-Help needed in understanding Improving Deep Neural Networks: Hyperparameter tun	4	646	February 22, 2022
Intuition on weight initialization Improving Deep Neural Networks: Hyperparameter tun	1	524	November 1, 2022
Course 2 Week 1 PA 1: Why does zero weight cause no change in loss Improving Deep Neural Networks: Hyperparameter tun	2	631	January 5, 2022