Deep learning is where models stop relying on hand-crafted features and start learning them directly from raw data. Stacks of simple layers, trained together, can recognise images, understand speech, and power the language models you use every day.
You'll begin with the intuition behind neural networks, then see how they learn through gradient descent and backpropagation. From there the module introduces convolutional neural networks for image data, and finishes by building a small neural network engine from scratch so you understand exactly what frameworks like PyTorch are doing under the hood.
With these foundations in place, you'll be ready for the next module, NLP, LLMs & Generative AI, where deep networks learn to read, write, and generate language.
A visual introduction to neurons, layers, weights, and biases, using handwritten digit recognition as the running example.
What a cost function is and how repeatedly stepping downhill lets a network improve its weights from data.
An intuitive walk-through of how errors flow backward through the network to tell each weight how to change.
How filters, pooling, and fully connected layers combine to classify images, explained step by step.
Build a tiny neural network library from nothing, writing automatic differentiation and training a small network by hand.