Learn ML / Stage 4
4 · Deep learning
Neural networks, from first principles to transformers and language models. Builds on stages 1 to 3.
By the end of this stage, you can…build and train a neural network in PyTorch, explain backpropagation, choose between a CNN, a recurrent network and a transformer for a given kind of data, and read the methods section of a deep learning paper.
What you will learn
- Neural network basics Essential
- PyTorch and JAX Essential
- Convolutional networks Essential
- Sequence models: RNNs and LSTMs Essential
- Transformers and language models Essential
- Generative models Essential
- Graph neural networks Essential
- Interpretability If you need it
Resources per subject
Neural network basics Essential
- Michael Nielsen, Neural Networks and Deep LearningA free book that derives backpropagation step by step, with code. 3Blue1Brown's videos are a good visual companion.★ recommended · free book · code
- 3Blue1Brown, Neural networksvideos
- TensorFlow Playgroundinteractive · train a small network in your browser
10 more options
- The Neural Network Zoo (Asimov Institute)overview chart, 2016–2019 · still useful for the classic architectures, but has no transformers, diffusion models or graph networks
- Understanding Deep Learning (Prince, University of Bath)free book · notebooks
- MIT 6.7960: Deep Learning (graduate)videos · notes
- CMU 11-785: Introduction to Deep Learningvideos · slides · homework
- NYU Deep Learning, Yann LeCun & Alfredo Canzianivideos · notes · notebooks
- MIT 6.S191: Introduction to Deep Learningshort course · videos · labs
- Deep Learning: Foundations and Concepts (Bishop & Bishop)free online book
- Dive into Deep Learning (Zhang, Lipton, Li & Smola)free book · code
- Neural Networks: Zero to Hero, Andrej Karpathyvideos · build everything from scratch
- Practical Deep Learning, fast.aivideos · code
PyTorch and JAX Essential
- Learn PyTorch for Deep Learning (Bourke)Beginner-friendly book and videos. PyTorch is the framework most ML research and biology tools are built on.★ recommended · free book · videos
- PyTorch tutorials (official)tutorials · code
- UvA Deep Learning tutorials, University of Amsterdamnotebooks · PyTorch and JAX
1 more option
- JAX tutorials (official)tutorials · code
Convolutional networks Essential
- Stanford CS231n: Deep Learning for Computer VisionThe standard university course on CNNs, with excellent notes.★ recommended · notes · videos
Sequence models: RNNs and LSTMs Essential
- Dive into Deep Learning: recurrent neural networks and LSTMsTheory with runnable code, written by academics.★ recommended · free book chapters · code
- Understanding LSTM Networks (Olah)visual explainer
- The Unreasonable Effectiveness of RNNs (Karpathy)blog post
Transformers and language models Essential
- Stanford CS224N: NLP with Deep Learning (Manning)Builds from word vectors to transformers and large language models. Then CS336 to build a language model yourself.★ recommended · videos · notes
- Stanford CS336: Language Modeling from Scratchlectures · assignments
- Stanford CS25: Transformers Unitedseminar videos
3 more options
- The Illustrated Transformer (Alammar)visual explainer
- Andrej Karpathy, YouTube channelvideos · how large language models work
- Hugging Face coursestransformers · LLMs · code
Generative models Essential
- Stanford CS236: Deep Generative Models (Ermon)VAEs, GANs, normalising flows and diffusion, in one course.★ recommended · notes · slides
- Berkeley CS294-158: Deep Unsupervised Learningvideos · slides
Graph neural networks Essential
- Stanford CS224W: Machine Learning with Graphs (Leskovec)The standard course on learning from graphs; the basis for most molecular ML.★ recommended · videos · notes
Interpretability If you need it
- Christoph Molnar, Interpretable Machine LearningSHAP, feature importance and other ways to explain model predictions, which matters for biology. ARENA covers the mechanics inside networks.★ recommended · free book
- ARENA: Alignment Research Engineer Acceleratorcourse · notebooks