Neural networks
Stack neurons into layers and watch a network bend space to separate data. Step through backpropagation with real numbers, train networks in your browser, and see what a convolution detects.
- Level
- Intermediate
- Length
- 6 lessons, 1 hr 38 min
- Assumes
- The machine learning track, or familiarity with loss and gradient descent.
- Builds on
- How machines learn
- 1Layers are matrix multiplicationsA layer of neurons is one matrix multiplication plus a bias. Seeing it that way explains the shapes in every neural network, and why GPUs train them so fast.14 min
- 2Why nonlinearity mattersWithout activation functions, a deep network is no more powerful than a single layer. Watch layers stretch the plane and activations bend it until a straight line can separate the data.15 min
- 3BackpropagationHow a network finds out which way to nudge every one of its weights. Step through the chain rule on a small computational graph, one node at a time.18 min
- 4Training a networkPut layers, activations, backpropagation, and gradient descent together and train real networks in your browser. See what each setting changes, and what each neuron learns.20 min
- 5ConvolutionsImages need a different kind of layer: a small window of weights that slides across the picture, finding the same pattern wherever it appears. Try classic edge detectors and design your own.16 min
- 6Reading handwritingA real network, trained on 60,000 handwritten digits, running in your browser. Draw a digit and watch 784 pixels flow through two hidden layers into ten probabilities.15 min
Labs in this track
- A layer as a matrixSee one layer of neurons two ways at once: as a network diagram and as the matrix product Wx + b.
- Backpropagation, step by stepRun a single neuron forward, then send gradients backward through its computational graph one chain-rule multiplication at a time.
- Convolution explorerSlide a 3 × 3 kernel across an image and see the arithmetic behind each output pixel. Try edge detectors, blur, and your own kernels.
- Handwritten digit recognizerDraw a digit for a network trained on MNIST and watch its hidden layers and output probabilities respond as you draw.
- Neural network playgroundTrain real neural networks in your browser. Change depth, width, activation, and learning rate, and watch every neuron and the loss curves respond.
- Watch a network reshape spacePush a grid and a dataset through a tiny network, layer by layer, and see linear maps stretch the plane while activations bend it.