Notebooks

Use the rendered version for reading and the .ipynb file for running or modifying the code.

Foundations and training

Notebook	Lecture	Format / purpose
Probability models, MLE, and MAP in PyTorch	L1	HTML · Colab · IPYNB
Robust linear regression: Gaussian, Laplace, and Student-t	L1	HTML · Colab · IPYNB
Computation graphs, backpropagation, and autograd	L4	HTML · Colab · IPYNB
micrograd-mlp	L2–L4	scalar autodiff, backpropagation, and a small MLP
optimizers	L5	SGD, momentum, and Adam on the same loss surface
cnn-shape-tour	L8	convolutional tensor shapes, parameters, and receptive fields
debug-ladder	L6–L7	diagnose a training run with deliberately planted failures

Sequences and Transformers

Notebook	Best after	What it makes concrete
rnn-by-hand	L13	an RNN cell and backpropagation through time
attention-by-hand	L16	attention from matrix multiplications
nanogpt	L16–L17	train and sample from a small GPT

Representation and generation

Notebook	Best after	What it makes concrete
contrastive-toy	L18	contrastive representation learning on a toy dataset
vae-mnist	L20	the VAE objective and the KL ablation
ddpm-2d	L22–L23	the forward noising process and learned denoising

Short companions

Small, lecture-specific notebooks remain available as source downloads under notebooks/. Use them when the deck points to one; they are not rendered as separate public pages.

Before an assignment

Open the notebook that the assignment names, restart it, and run it top to bottom. A result you cannot reproduce or explain is not yet useful evidence for your submission.