Neural Networks & Computer Vision
Deep learning from scratch — perceptrons, backprop, CNNs, RNNs, transformers, and hands-on projects.
📎 Assignments & course material (original docs shared in the course)
Why deep learning quietly took over the world
⏱ 19 min read · 📖 4 topics
Welcome to your very first Neural Nets class.
Why "deep" actually wins — a second, deeper look
⏱ 20 min read · 📖 4 topics
In Session 1 we met the headline idea: stack many layers and the network learns its own features.
From Perceptron to Multi-Layer Perceptron
⏱ 20 min read · 📖 4 topics
Welcome back!
How a neural network actually learns
⏱ 20 min read · 📖 4 topics
This is the session where the magic stops being magic.
How a neural net measures its mistakes — and fixes them
⏱ 21 min read · 📖 5 topics
In Session 4 we learned how backpropagation figures out which way each weight should move.
Regularization — Fighting Overfitting
⏱ 21 min read · 📖 5 topics
Your network can train perfectly on its homework and still flunk the real exam.
Why training a deep net sometimes just… fails (and how to fix it)
⏱ 23 min read · 📖 5 topics
You've built a network, you've picked a loss, you press "train" — and the loss either sits flat forever or explodes into NaN .
Let's actually build a neural network — twice
⏱ 20 min read · 📖 4 topics
Up to now we've been learning the pieces: neurons, activations, gradient descent, backprop, regularization, initialization.
Convolutions — Seeing with Neural Networks
⏱ 23 min read · 📖 5 topics
Welcome to the start of the Computer Vision part of the course.
The famous CNNs, from a tiny digit reader to 152 layers deep
⏱ 19 min read · 📖 5 topics
Last session you learned what a convolution is .
Standing on the shoulders of giant models
⏱ 18 min read · 📖 4 topics
In Session 10 we built powerful CNN architectures.
Teaching a network to remember: how RNNs read things in order
⏱ 20 min read · 📖 5 topics
Until now our networks looked at one thing at a time — one image, one fixed row of features.
Teaching a network to remember (and forget) on purpose
⏱ 21 min read · 📖 5 topics
Last session we built plain RNNs and watched them choke on long sequences — the dreaded vanishing gradient.
Teaching a network to translate — and to look back
⏱ 17 min read · 📖 4 topics
Last few sessions we built up RNNs, LSTMs and GRUs — networks that read sequences one step at a time.
The Transformer, from the dot product up
⏱ 21 min read · 📖 5 topics
Last session we taught a recurrent network to translate by reading one word at a time and leaning on an attention "spotlight." Today we take a bold…
BERT, GPT & how machines chop text into tokens
⏱ 21 min read · 📖 5 topics
Last session we built the Transformer from scratch.
Turn your photo into a Van Gogh painting
⏱ 19 min read · 📖 4 topics
This is a project week — we build something real and beautiful from scratch.
Build a Shakespeare bot and a translator
⏱ 21 min read · 📖 4 topics
This is a project week — less lecture, more building.
Let's build a tiny GPT, from scratch, that writes text
⏱ 22 min read · 📖 4 topics
This is a project session — no new theory dumped on you, just rolling up our sleeves and building a real, runnable mini-Transformer in PyTorch .
Project Week: train your very own small language model
⏱ 19 min read · 📖 4 topics
This is the capstone — the session where everything you've learned becomes a real, running thing.