Class meeting for 10-605 Deep Learning

This is one of the class meetings on the schedule for the course Machine Learning with Large Datasets 10-605 in Fall_2016.

Slides

• Lecture 1: Powerpoint, PDF.

Readings

• Automatic differentiation:
  • William's notes on automatic differentiation, and the Python code for a simple Wengart list generator and a sample use of a one.
  • Domke's blog post - clear but not much detail - and another nice blog post.
  • The clearest paper I've found is Reverse-Mode AD in a Functional Framework: Lambda the Ultimate Backpropagator

• More general neural networks:
  • Neural Networks and Deep Learning An online book by Michael Nielsen, pitched at an appropriate level for 10-601, which has a bunch of exercises and on-line sample programs in Python.
  • For much much more detail, look at the MIT Press book (in preparation) from Bengio - it's very complete but also fairly technical.

Things to remember

• The underlying reasons deep networks are hard to train
• Exploding/vanishing gradients
• Saturation
• The importance of key recent advances in neural networks:
• Matrix operations and GPU training
• ReLU, cross-entropy, softmax
• How backprop can be generalized to a sequence of assignment operations