Machine Learning & Training

What Is Weight Initialization?

Weight initialization refers to how the initial values of a neural network's parameters are chosen before training. Good initialization helps signals and gradients flow properly through the network and can speed up convergence. Poor initialization may lead to vanishing or exploding gradients.