|
Something that assumes you're basically competent, mathematically and statistically. |
|
I would suggest Peter Hinton's Google talk as a starter, and his more recent video lecture on Deep Belief Networks for more detail. Hinton's website has papers, references, and code to further your studies. Finally, If you prefer a written introduction, Learning Deep Architectures for AI provides a good overview of the motivations and methods of DBNs. 1
I would second the recommendation for reading "Learning Deep Architectures for AI", Yoshua Bengio's survey tutorial. Minor correction: I think you are interested in deep architectures in general, not DBNs (the first general purpose deep architecture).
(Jul 01 '10 at 15:03)
Joseph Turian ♦♦
|
|
If you are interested in trying a deep method on an NLP task, you could try implementing a deep CRF: Yu et al (2010) "Language Recognition Using Deep-Structured Conditional Random Fields" (or you could ask these researchers for their implementation). |
|
Once you've watched Hinton's Google Talk as a general introduction as suggested by John, I recommend the hands-on approach to get your first DBN up and munging MNIST digits on your favorite GPU or CPU by following the marvelous deep-learning tutorial (you can skip the denoising autoencoders and convolutional neural nets steps if you want to focus on DBNs). |
how about geoffrey hinton's tutorial at videolectures.net: http://videolectures.net/jul09_hinton_deeplearn/