Weight Initialization in Neural Network: Part 2

In Part 1, we said good initialization keeps the signal at a stable scale across all layers, and bad initialization either explodes or… Continue reading on Medium »