Explainers
Learn
The ideas behind the models, one at a time. Each starts with a short animation, explains the idea in a few paragraphs and ends with a lab to play with. The Models pages show what each idea achieves once trained; these show why it works.
- 01
Training
Backpropagation
How does a wrong guess turn into better weights?
14 sanimation, then a lab
used in MLP in NumPy
Interactive
- 02
Vision
Convolution
Why does one small filter beat a weight for every pixel?
14 sanimation, then a lab
used in Small CNN
Interactive
- 03
Vision
Skip connections
Why does a deeper network train worse, and what does a shortcut fix?
8 sanimation, then a lab
used in ResNet on CIFAR-10
Interactive
- 04
Sequence
Attention
How does a transformer decide which earlier words matter?
9 sanimation, then a lab
used in Mini-GPT
Interactive
- 05
Generative
Latent space
What does a model need so that every point between two digits makes sense?
11 sanimation, then a lab
used in Variational autoencoder
Interactive
- 06
Training
Overfitting
Why can a model be right on its training data and wrong on new data?
9 sanimation, then a lab
used in ViT-Tiny
Interactive