What is the purpose of weight initialization in deep networks?

Question

MLQuiz · Accepted Answer

Proper initialization (e.g., Xavier, He) prevents activations from becoming too small or too large, aiding convergence and avoiding vanishing/exploding gradients.