Probabilistic Models for Predicting Multiple Correlated Labels

Data Mining SIG

Title: Probabilistic Models for Predicting Multiple Correlated Labels

Speaker: Kevin Murphy

Event Details
Joint probability models are useful for unsupervised learning (“knowledge discovery”) as well as supervised learning tasks such as where we want to predict multiple correlated outputs (“collective classification”). One approach to representing such models is to explicitly model the correlations using a graphical model, where we add edges between correlated variables (nodes). Another approach is to implicitly model the correlations via a set of latent common factors; this induces dependence between the visible variables without the need to add explicit edges between them. In this talk, I compare both of these methods, using as a running example the task of predicting tags for images.

Speaker Bio
Kevin Murphy is a research scientist at Google, where he is currently working on problems related to entity extraction from text and video. He is on leave from the University of British Columbia in Vancouver, Canada, where he is an associate professor in the departments of computer science and statistics. Kevin jointed UBC in 2004. Prior to that, Kevin did a postdoc at MIT, his PhD at UC Berkeley, his MSc at U. Pennsylvania, and his BSc at U. Cambridge in England. Kevin is best known for his work in the area of Bayesian networks/ graphical models. He has just finished writing a textbook called “Machine learning: a probabilistic perspective”, to be published by MIT Press in August 2012.

Event page provided by ACM

http://www.sfbayacm.org/event/probabilistic-models-predicting-multiple-correlated-labels