Practical Deep Learning for Coders by fast.ai has you training a state-of-the-art image classifier in the first lesson, then spends the rest of the course explaining why it works. That top-down approach suits people who learn by doing, and Jeremy Howard is candid about which details matter and which are academic. It's free, uses PyTorch, and the accompanying book is free online too. You need a year or so of Python; you don't need a maths degree.
Intuition-first videos with practical projects. Gentler pacing than fast.ai.
Long, carefully referenced surveys of attention, diffusion, and agents. Better than most review papers.
Lilian Weng
The modern textbook: clear figures, current architectures, and free. Our pick over older references.
Simon J.D. Prince
The foundational textbook. Parts are dated, but its math chapters remain a reference.
GBGoodfellow, Bengio & Courville
Short courses on current tools like RAG and agents, often taught with the toolmakers.
DeepLearning.AI·1–2 hours each
Transformers, fine-tuning, and datasets with the library everyone uses. Free and practical.
Hugging Face
The standard NLP course, now centered on transformers and LLMs. Lectures are free online.
Stanford's legendary vision course. The assignments make you implement everything yourself.
An interactive textbook with runnable code. Theory and practice on the same page.
Dive into Deep Learning
Official, runnable tutorials. Start with "Learn the Basics" before any third-party course.
PyTorch
Search, optimization, and neural networks in Python. A rigorous tour before specializing.
Andrew Ng's bottom-up counterpart to fast.ai. Clear on the math behind each architecture.
Build backprop, then a GPT, from scratch. The best way to really understand LLMs.
Andrej Karpathy
Visual intuition for backpropagation, transformers, and attention. Watch before starting any course.
3Blue1Brown
No picks match this filter.
Course availability, level, and pricing are set by the provider and can change. Verify on the provider's site before enrolling.
