BEGIN:VCALENDAR VERSION:2.0 PRODID:-//Talks.cam//talks.cam.ac.uk// X-WR-CALNAME:Talks.cam BEGIN:VEVENT SUMMARY:Spectral Clustering meets Graphical Models - Yuri Boykov (Universi ty of Western Ontario) DTSTART:20170904T101000Z DTEND:20170904T110000Z UID:TALK77611@talks.cam.ac.uk CONTACT:INI IT DESCRIPTION:Co-authors: Dmitri Marin (UWO)\, Meng Tang (UWO)\, Ism ail Ben Ayed (ETS\, Montreal)

This talk discusses t wo seemingly unrelated data analysis methodologies: kernel clustering and graphical models. Clustering is widely used for general data where kernel methods are particularly popular due to their discriminating power. Graphi cal models such as Markov Random Fields (MRF) and related continuous geome tric methods represent the state-of-the-art regularization methodology for image segmentation. While both clustering and regularization models are v ery widely used in machine learning and computer vision\, they were not co mbined before due to significant differences in the corresponding optimiza tion\, e.g. spectral relaxation vs. combinatorial methods for submodular o ptimization and its approximations. This talk reviews the general properti es of kernel clustering and graphical models\, discusses their limitations (including newly discovered "density biases" in kernel methods)\, and pro poses a general unified framework based on our new bound optimization algo r ithm. In particular\, we show that popular MRF potentials introduce prin cipled geometric and contextual constraints into clustering\, while standa rd kernel methodology allows graphical models to work with arbitrary high- dimensional features.

Related Links

https://arxiv.org/abs/1506.07439 - bound optimizatio n for kernel clustering
https://arxiv.org/abs/1705.05950 - about density biases in kernel clus tering

LOCATION:Seminar Room 1\, Newton Institute END:VEVENT END:VCALENDAR