A neural network is a function built from many small units called neurons. Each neuron takes some numbers in, multiplies them by weights, adds a bias and passes the result through an activation function. Stack neurons into layers and the network can turn raw input, such as the pixels of a handwritten digit, into a useful prediction.
This module builds that picture up from a single neuron. You will see why layers matter, how information moves from input to output in the forward pass, and why the weights and biases are the "knobs" that a network learns. How those knobs get tuned is the subject of the next module.
Watch the lessons in order, then continue to Backpropagation & Gradient Descent.
3Blue1Brown introduces neural networks through handwritten digit recognition: what the neurons are, why there are layers, and the math behind weights, biases and the connections between layers.
Video 18minStatQuest with Josh Starmer opens up the "black box": activation functions, hidden layers, and how a network bends and reshapes simple curves to fit data points and make predictions.
Video 18minAssemblyAI on why activation functions are needed and how the common ones differ: step, sigmoid, tanh, ReLU, Leaky ReLU and softmax, with a look at how to use them in code.
Video 6min