Unfortunately, a scheduling failure meant I missed all of AIStat and most of the learning workshop, otherwise known as Snowbird, when it’s at Snowbird.
At snowbird, the talk on Sum-Product networks by Hoifung Poon stood out to me (Pedro Domingos is a coauthor.). The basic point was that by appropriately constructing networks based on sums and products, the normalization problem in probabilistic models is eliminated, yielding a highly tractable yet flexible representation+learning algorithm. As an algorithm, this is noticeably cleaner than deep belief networks with a claim to being an order of magnitude faster and working better on an image completion task.
Snowbird doesn’t have real papers—just the abstract above. I look forward to seeing the paper. (added: Rodrigo points out the deep learning workshop draft.)
What about this paper from the same authors ?
“Sum-Product Networks: A New Deep Architecture”
http://deeplearningworkshopnips2010.files.wordpress.com/2010/11/main.pdf
Right, thanks. I added it.
An extended paper and source code are now available: http://alchemy.cs.washington.edu/spn/
The source code has some nice optimization tricks not mentioned in the paper.
For example, it uses the knowledge of the specific network structure. It is then able to find the max-valued product nodes faster.
If you want to see examples of simple sum-product networks,
I put them on my blog:
http://lessoned.blogspot.com/2011/10/intro-to-sum-product-networks.html