⛰️
Learning = going downhill
Loss, gradients, learning rate and backpropagation, explored in 3D.
15 min 90 XP🧠 Thinkers
Step 1 of 11
A network starts with random weights, so its first predictions are terrible. How can we find good values among billions of parameters? We need two ingredients: a way to measure how wrong we are, and a way to improve.