Berk Kapicioglu will present his preFPO on Wednesday February 22 at 11AM in Room 402. The members of his committee are: Rob Schapire, advisor; David Blei and Tony Jebara (Columbia), readers; Jennifer Rexford and Andrea LaPaugh, nonreaders. Everyone is invited to attend his talk. His abstract follows below. ----- Original Message ----- Title: Applications of Machine Learning to Location Data. Abstract: First, we introduce a new graphical model for tracking radio-tagged animals and learning their movement patterns. The model provides a principled way to combine radio telemetry data with an arbitrary set of user-defined, spatial features. We describe an efficient stochastic gradient algorithm for fitting model parameters to data and demonstrate its effectiveness via asymptotic analysis and synthetic experiments. We also apply our model to real datasets, and show that it outperforms the most popular radio telemetry software package used in ecology. We conclude that integration of different data sources under a single statistical framework, coupled with appropriate parameter and state estimation procedures, produces both accurate location estimates and an interpretable statistical model of animal movement. Second, we present the “Place Discovery” problem, where the goal is to provide a context-dependent and personalized ranking of nearby places. To the best of our knowledge, we are the first to pose this as a machine learning problem. We provide a new matrix approximation algorithm that leverages implicit bipartite ranking and spatial constraints. We demonstrate on real location datasets that our methods significantly improve on the amount of personalization achieved by state-of-the-art matrix approximation methods, such as Maximum-Margin Matrix Factorization. Lastly, we talk about “Location Prediction”. Here, the goal is to predict the coordinates (i.e. latitude and longitude) of a user at an arbitrary time in the future. We design an unsupervised hierarchical Bayesian graphical model for location prediction that learns natural topics such as home and work from unlabeled location data. We apply our model both to data gathered from mobile applications, where there are many users but each user has sparse location points, and to data gathered from cellular carriers. Our model outperforms existing methods in both cases.