Fastai Study Group

Mar 2, 2019 • Thomas Tumiel

Starting in March, CryptoSoc is leading a study group! Fastai is an online course that teaches you how to do practical deep learning starting from only high-school maths and a bit of programming knowledge. Over the course of 7 weeks, there will be numerous topics covered:

  • Computer vision
  • Gradient descent
  • Multi-label classification
  • Image segmentation
  • Natural language processing
  • Tabular data processing
  • Backpropogation
  • PyTorch, GANs, CNNs, using GPUs and more!

The course is a comprehensive introduction to state-of-the-art deep learning and focuses on a top-down, practical approach.

Currently around 50 society members have joined the study group. The lectures will be watched individually at the start of the week and then students are encouraged to work through the original notebook and then reimplement the code from scratch. We’ve created a repo that has stripped down versions of the weekly notebooks that will help students learn if they find themselves struggling. Additionally, throughout the week, we will be communicating on Slack whereby members can ask questions and receive feedback from both the group leaders and the other students. In-person group study sessions will be held on UCT campus so that we can ideate, prototype and debug together. Finally, we will hold a few Kaggle competitions to test out these new skills in a competitive way!

This is a very exciting course and we are very hopeful about the participants and all the amazing things they will go forth to create.