University of Cambridge > > Machine Learning @ CUED > Sum-Product Networks for Probabilistic Modeling

Sum-Product Networks for Probabilistic Modeling

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Louise Segar.

In machine learning and artificial intelligence, probabilistic graphical models are a principled and widely used approach for dealing with uncertain knowledge. However, one of their downsides is that exact inference quickly becomes intractable in practical models. Therefore, one often uses simple models, allowing exact inference, which however potentially undermodel a given problem, or one uses approximate inference methods, whose performance is often hard to assess for particular applications. Sum-Product networks (SPNs) are a new avenue for probabilistic modeling, promising a remedy for this problem: Using a deep network structure, they are able to represent highly complex variable dependencies, while at the same time many inference scenarios can be solved with computational costs linear in the representation size of the SPN . In this talk, I give an introduction to SPNs, and discuss basic notions such as completeness, consistency and decomposability, which enable tractable inference. I further present some recent theoretical and practical results, together with their implications on representational properties, learning and inference. Finally, I present some results of SPNs applied to computer vision and speech modeling.

This talk is part of the Machine Learning @ CUED series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.


© 2006-2024, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity