Back-propagation is the algorithm of choice for learning in neural networks and the deep nets. Though it is simple to understand, one needs to learn some more tricks to make it work for multi-layered network. There are some choices which can be made to speed up learning. These slides summarize few of these tricks.

References

  1. "Learning representation by back-propagating errors" Link
  2. "Efficient Backprop" Link
  3. Neural network and deep learning Link