Go beneath the surface of modern AI. These courses teach you to build and train neural networks from scratch and with industry frameworks, from top-down practical approaches to rigorous university treatments.
Sorted cheapest first. Click a column header to re-sort. Programmers ready for the deep end.
| Course | Level | Duration | Price | Certificate | |
|---|---|---|---|---|---|
d2l.ai (Zhang, Lipton, Li & Smola) · d2l.ai |
intermediate | Self-paced (full textbook) | Free | No | Visit |
MIT · MIT OpenCourseWare |
intermediate | 9 lectures plus 3 software labs (~2 weeks intensive) | Free | No | Visit |
3Blue1Brown (Grant Sanderson) · YouTube |
beginner | ~3 hours total (8 videos) | Free | No | Visit |
NYU Center for Data Science · GitHub |
advanced | 15 weeks of lectures and practica | Free | No | Visit |
fast.ai |
intermediate | 9 lessons, ~90 minutes each plus notebooks | Free | No | Visit |
DeepLearning.AI · Coursera |
intermediate | 3 months at 10 hrs/week (~129 hours, 5 courses) | Free to audit Free to audit · $49/mo |
Yes | Visit |
DeepLearning.AI · Coursera |
intermediate | 2 months at 10 hrs/week (~83 hours, 4 courses) | Free to audit Free to audit · $49/mo |
Yes | Visit |
Udacity |
intermediate | ~50 hours (about 2-3 months part-time) | Paid $249/mo |
Yes | Visit |
No payment required. Several of these are complete university courses.
An interactive, code-first textbook with every concept implemented in PyTorch, NumPy/MXNet, JAX, and TensorFlow, from linear regression through transformers and optimization. Best for learners who want a rigorous reference that doubles as a hands-on course.
MIT's fast-moving bootcamp updates each year to cover sequence modeling, vision, generative models, RL, and LLMs, with TensorFlow/PyTorch labs including LLM fine-tuning. Great for getting current on the deep learning landscape in a compact format.
Grant Sanderson's animated series builds visual intuition for what neural networks compute, how gradient descent and backpropagation work, and how transformers and attention process text. Watch it before or alongside any formal course to make the math click.
Yann LeCun's lectures paired with Alfredo Canziani's PyTorch practica cover energy-based models, self-supervised learning, graph networks, and more, with notebooks and multilingual notes. For learners who want a research-oriented perspective straight from a founder of the field.
Jeremy Howard's top-down course has you training state-of-the-art image, text, and tabular models in lesson one, then peels back the layers to the math. Ideal for working programmers who want results first and theory second.
Watch everything free; pay only if you want graded assignments and a credential.
Five courses covering neural network foundations, hyperparameter tuning, ML strategy, CNNs, and sequence models and transformers, taught by Andrew Ng. The default next step after an intro ML course for anyone heading toward deep learning roles.
Laurence Moroney's four-course track teaches building CNNs, NLP models, and time-series forecasters in TensorFlow and Keras. Pick it if your target job stack is TensorFlow or you want applied practice after a theory-heavy course.
Structured programs with instructors, cohorts, projects or formal credentials.
Project-reviewed program covering neural networks, CNNs, RNNs, GANs, and deployment with PyTorch, with human feedback on each submitted project. Worth it if you want a portfolio of graded projects and mentor support rather than videos alone.
Supervised and unsupervised learning, model evaluation, and the math behind it.
Image classification, object detection, segmentation, and generative image models.
Text processing, embeddings, sequence models, and modern transformer NLP.
Transformers, large language models, fine-tuning, and building with foundation models.