Sequence Models and Generative Models

reading · we frame, you read MIT or the canonical taught · we author, no canonical fits ↺ spirals back to earlier lessons

Course locked

Complete Computer Vision first.

This course unlocks once you've finished its prerequisite. Open prerequisite →

Prerequisites

Linear Algebra not yet

Neural Networks and Deep Learning Fundamentals not yet

Complete prerequisites first to enroll.

reading · 35 min + ~240 min canonical

Recurrent Networks, LSTMs, and the Memory Problem

CNNs (last course) conquered images — data with spatial structure. But a huge fraction of the world's data is sequential: text (a sequence of words), speech (a sequence of sounds)…

primary Andrew Ng Deep Learning Specialization — Course 5: Sequence Models

preview →

reading · 35 min + ~240 min canonical L1

Attention and the Sequence-to-Sequence Revolution

The last lesson left LSTMs with two limits: they process sequentially (slow, unparallelizable) and still struggle with long ranges. This lesson introduces the idea that began to f…

primary Andrew Ng Deep Learning Specialization — Course 5: Sequence Models

after L1 locked

reading · 40 min + ~240 min canonical L1

The Transformer: Attention Is All You Need

This is the most important architecture in modern AI — the one behind every large language model (ChatGPT, Claude, Gemini), the one that conquered vision too (ViT, last course), t…

primary Andrew Ng Deep Learning Specialization — Course 5: Sequence Models

after L2 locked

reading · 35 min + ~240 min canonical L1

Transformer Variants: BERT, GPT, and the Models That Changed Everything

The transformer (last lesson) is one architecture — but it comes in three flavors, and the difference between them explains the whole landscape of modern language AI, including wh…

primary Andrew Ng Deep Learning Specialization — Course 5: Sequence Models

after L3 locked

reading · 40 min + ~240 min canonical L1

Generative Models: VAEs, GANs, and Diffusion

Everything in this course so far — and most of classical ML — has been discriminative: given an input, predict a label ("what is this?"). Generative models do something profoundly…

primary Andrew Ng Deep Learning Specialization — Course 5: Sequence Models

after L4 locked

exercise reading · 15 min L1

Exercises: Easy (Warm-up)

- [ ] Implement an RNN cell from scratch in PyTorch — Single timestep forward, concatenate hidden state + input - [ ] Train a character-level LSTM on Shakespeare text — Generate n…

after L5 locked

exercise reading · 30 min L1

Exercises: Medium (Core)

- [ ] Karpathy Lecture 7: Build GPT from scratch — Transformer decoder: token/position embeddings, multi-head attention, feed-forward, train on Shakespeare - [ ] Karpathy Lecture …

after L6 locked

exercise reading · 45 min L1

Exercises: Hard (Stretch)

- [ ] Implement a full Transformer (encoder-decoder) for machine translation — English to French on a small parallel corpus - [ ] Train a diffusion model on MNIST — Forward noisin…

after L7 locked

exercise reading · 60 min L1

Mini-Project

- [ ] Build a story generator — Train a small GPT model (following Karpathy) on a text corpus of your choice; implement temperature sampling, top-k, top-p; build a simple web UI f…

after L8 locked

L10

project reading · 480 min L1

Project: Build GPT from Scratch

Follow Karpathy's lecture 7. Implement a decoder-only Transformer: token embeddings, positional embeddings, multi-head self-attention, feed-forward layers, layer norm. Train on Sh…

after L9 locked

L11

project reading · 240 min L1

Project: BPE Tokenizer

Follow Karpathy's lecture 8. Implement byte-pair encoding from scratch: start with bytes, iteratively merge most frequent pairs, build vocab. Encode and decode text. Compare token…

after L10 locked

L12

assessment reading · 56 min L1

Self-Assessment: Sequence Models and Generative Models

- [ ] Explain the Transformer architecture: self-attention, multi-head attention, positional encoding, feed-forward layers. - [ ] What is scaled dot-product attention? Why do we d…

after L11 locked

L13

assessment reading · 60 min L1

Quarter 15 Gate Check: Deep Learning

- [ ] Course 15.1 knowledge check passed - [ ] Course 15.2 knowledge check passed - [ ] Course 15.3 knowledge check passed - [ ] Train and Deploy Neural Networks capstone completed

after L12 locked

13 lessons. Read in order; spiral back when you need to. By the end you'll have used the core ideas twice — once on the abstract, once on something you'll meet at work next week.