Follow the 14-module curriculum to learn deep learning from Python basics through GANs and reinforcement learning at no cost.
Use individual notebooks as reference material when implementing a specific technique like convolutional networks or time series models.
Run the course notebooks in Google Colab without installing anything locally to learn deep learning hands-on.
This is the older TensorFlow/Keras edition, the university's current course uses PyTorch in a separate repository.
This repository contains the course materials for T81-558, a graduate-level class on deep learning taught by Jeff Heaton at Washington University in St. Louis. The materials are published openly on GitHub so anyone can follow along, not just enrolled students. The course covers how to build and train neural networks using Python, focusing on practical applications rather than heavy mathematics. Topics progress from Python basics through common neural network architectures: standard networks for tabular data, convolutional networks for image recognition, recurrent networks for time series, generative adversarial networks (GANs) that can produce new images or synthetic data, and reinforcement learning where a program learns by trial and error. The primary tools used are TensorFlow and Keras, which are software libraries that simplify the process of defining and training neural networks. An important note at the top of the README: the university now runs a newer version of this course that uses a different library called PyTorch instead of TensorFlow and Keras. That newer version lives in a separate repository. This repository is the older Keras/TensorFlow edition, which is no longer the active course offering but remains publicly available. Students currently enrolled at Washington University should use the PyTorch version. The materials exist as Jupyter Notebooks, which are interactive documents that mix explanatory text with runnable code. Each module in the course has its own notebook. The full course content is also available as a printed textbook through the instructor's website, and all notebooks can be run in Google Colab, a free cloud-based environment, so you do not need to install anything locally to work through the material. The syllabus in the README covers 14 modules across a semester, with a final project and two in-person meeting days for the hybrid format. The course is aimed at people who have some programming experience but do not need prior knowledge of Python or neural networks.
← jeffheaton on gitmyhub — every repo by this author, as a profile.
Verify against the repo before relying on details.