We make short, AI-voiced lessons. Before a video ships, every sentence goes through automatic checks, and then a person listens. The automatic checks kept failing in one place: names. This post explai
Deep Learning, Visually #20 Watch the Short: youtube.com/shorts/yYxpjlvW69s Why the numbers drift A deep network is a chain of layers. Each layer takes a weighted sum of the activations from the layer
Vanishing and exploding gradients: why blame fades or blows up on its way back through a deep network, and the fixes that made deep learning work
Weight initialization: why a network's dials must start different, and how big the random start should be
Learning-rate schedules: start big, halve the step size every so often, and settle where one step size can't
Adam, step by step: why one shared step size fails, and how giving every dial its own fixes it
Momentum, step by step: give gradient descent a memory, and it rolls past the traps that stop it
Stochastic and mini-batch gradient descent: why every real training loop takes small, noisy steps
The learning rate, step by step: why one number decides whether training works at all