Raquel Urtasun is an Assistant Professor at TTI-Chicago a philanthropically endowed academic institute located in the campus of
the University of Chicago. She was a visiting professor at ETH Zurich during the spring semester of 2010. Previously, she was a postdoctoral
research scientist at UC Berkeley and ICSI. Before that, she was a postdoctoral associate at the Computer Science and Artificial
Intelligence Laboratory (CSAIL) at MIT, where she worked with Trevor Darrell. Raquel Urtasun completed her PhD at the Computer
Vision Laboratory, at EPFL, Switzerland in 2006 working with Pascal Fua and David Fleet at the University of Toronto. She has been area
chair of NIPS 2010 and 2011, she will be area chair of UAI 2012, and served in the committee of numerous international computer vision and machine learning conferences (e.g.,
CVPR, ICCV, ECCV, ICML, NIPS) as well as major journals (e.g., TPAMI, IJCV, JMRL). Her major interests are statistical learning and computer vision, with a particular interest in non-parametric Bayesian statistics, latent variable models, structured prediction and their application to 3D scene understanding and human pose estimation.

AG 1, AG 2, AG 3, AG 4, AG 5, RG1, SWS, MMCI   Info

AG Audience

English

60 Minutes

Saarbrücken

Supervised learning problems often involve the prediction of complex
labels such as sequences, trees or more complex structures.
Introducing dependencies between sets of labels is often required in
order to achieve high accuracy in these tasks. Unfortunately, this
results in prediction problems that are NP hard. In this talk I'll
present our recent work on approximated inference and learning for
structure prediction problems.
I'll first present how duality and belief propagation can be
integrated into efficient methods for recovering the maximum
a-posteriori (MAP) assignment, including hybrid graphical models with
mixtures of discrete and continuous variables. Importantly, these
algorithms can take advantage of parallel architectures while
maintaining their theoretical guarantees.
I will also show how duality and local inference can be applied to
approximate a family of structure prediction problems including
structured support vector machines (SVMs) and conditional random
fields (CRFs). I'll
demonstrate the effectiveness of these approaches in computer vision
problems. In particular, our efficient primal-dual solvers currently
achieve state-of-the-art performance in the problems of 3D indoor
scene understanding, semantic segmentation and 3D reconstruction from
stereo imagery.

Mario Fritz

+49 681 9325 1204

--email hidden

passcode not visible

logged in users only