The TensorFlow notebooks Nvidia once used in London

What it does This repo is a set of five interactive Jupyter notebooks that teach TensorFlow through examples and visualizations. The labs walk through feed-forward networks, convolutional nets with spatial transformers, recurrent encoder-decoder models with attention, autoencoders, and a Kaggle leaf-classification competition. Most of the coursework is translated from older Theano/Lasagne labs with minor modifications.

The interesting bit Labs one through four were actually used in a London “Deep Learning using TensorFlow” event run by Nvidia and Persontyle. That gives the curriculum a classroom pedigree rare among self-published notebook collections.

Key highlights

• Five structured labs progressing from logistic regression to autoencoders
• Includes a practical Kaggle “Leaf Classification” competition lab
• Covers attention-based encoder-decoder translation and Spatial Transformer Networks
• Runs on a laptop CPU, with optional GPU acceleration
• Points to solid external reading from Michael Nielsen, Alex Graves, and the Deep Learning Book

Caveats

• Four of the five labs are ported from Theano/Lasagne, so the TensorFlow patterns may carry older-framework baggage
• The README never states which TensorFlow version the notebooks expect

Verdict A solid starting point for TensorFlow newcomers who want structured, event-tested coursework spanning toy problems and a real competition. Look elsewhere if you need modern Keras idioms or PyTorch equivalents.