Home / Skills tracks / Neural networks and CNNs
At schoolPost 16 and adults

Neural networks and CNNs

Build a neuron, train it with gradient descent, then see how convolution lets a network read an image. Real code, no black boxes.

4 units, about 4 hours. Free.

1

A neuron is multiply, add, bend

Compute a neuron output by hand and in code, and explain why the activation function matters.

2

Training: loss and gradient descent

Explain how a loss function and gradients drive learning, and train a model in code.

3

Convolution: how a network sees

Apply a convolution kernel by hand and in code, and explain why CNNs beat dense layers on images.

4

Overfitting and honest evaluation

Detect overfitting from training and validation curves, and evaluate a model honestly.