Principles of the course/book:
  1. Teach the whole game (i.e. start with complete high-level approach, then dig down into abstractions)
  1. Teach through example
  1. Simplify
  1. Remove barriers


  1. Aim to solve your own problems
  1. Focus on your hobbies / passions
  1. Start small
  1. Encourage your playfulness / curiosity


Supervised ML paradigm:
notion image

Be aware!

What is the single most important and challenging issue when training? Reducing overfitting
What is the most important method for training models more quickly, accurately and with less data? Using pre-trained models / transfer learning
What is the single biggest source of failure when organisations decide to use AI? Poor understanding of importance of test / validation sets

Interpreting a neural network

How to interpret a (trained) neural network:
notion image
Left: reconstructed weights; Right: parts of training images with largest activations

Image Recognizers for Non-Image Tasks

How to use an image recogniser for a non-image task? Find a way of transforming the data into an image, such that the human eye can spot classes.


How to get FastAI documentation within a Jupyter notebook? Call doc(object)