// the find
jeffheaton/t81_558_deep_learning
T81-558: Keras - Applications of Deep Neural Networks @Washington University in St. Louis
This is Jeff Heaton's Washington University course on applying deep learning with Keras/TensorFlow — CNNs, LSTMs, GANs, transformers, reinforcement learning, NLP, and deployment, all as a numbered sequence of Jupyter notebooks. It's for self-learners who want a structured path through classic-to-2023-era deep learning rather than a library or production tool, and the README itself points current students to a newer PyTorch-based sibling repo.
The module sequence is genuinely well-organized — each notebook builds on the last and covers ground most tutorials skip, like OpenAI Gym reinforcement learning, StyleGAN3 training, and Flask model deployment, not just image classification. It's backed by an actual textbook (ISBN cited) and an arXiv paper, so the material has been reviewed and tested against real students over several semesters. Environment setup is explicit, with separate tensorflow.yml / tensorflow-gpu.yml / environment.yml conda files rather than a vague 'pip install the usual suspects'.
The author's own first line in the README tells you this repo is superseded — the live course moved to a PyTorch version, so this is frozen at Spring 2023 content and won't get fixes. Pinned Keras/TensorFlow and HuggingFace API usage from that era is already stale enough to need patching before notebooks run clean on current TF releases. There's no CI and no tests of any kind, so broken notebooks (dead dataset URLs, deprecated API calls) will just sit there until someone notices manually. A fair amount of the repo is university logistics — assignment templates, grading rubrics, syllabus PDFs — that's dead weight for anyone using this outside the actual class.