BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Sparsity pattern aggregation for convex stochastic optimization. -
  Phillipe Rigollet (Princeton University)
DTSTART:20101029T150000Z
DTEND:20101029T160000Z
UID:TALK25760@talks.cam.ac.uk
CONTACT:Richard Nickl
DESCRIPTION:Important statistical problems including regression\, binary c
 lassification\nand density estimation can be recast as convex stochastic o
 ptimization\nproblems when seen from the point of view of statistical aggr
 egation. These\nconvex problems can be numerically solved efficiently in h
 igh dimension but\nmay show mediocre statistical performance. One way to o
 vercome this\nsituation consists in assuming that there exists approximate
  solution\,\ncalled "sparse"\, that are of moderate dimension. This presen
 tation\nintroduces a new method called "exponential screening (ES)" as an\
 nalternative to the $\\ell_1$-penalization idea\, which is currently the m
 ost\npopular way to find these sparse solutions. While $\\ell_1$ based met
 hods can\nbe analyzed only under rather stringent assumptions\, ES shows o
 ptimal\nstatistical performance under fairly general assumptions. Implemen
 tation is\nnot straightforward but it can be approximated using the Metrop
 olis\nalgorithm which results in a stochastic greedy algorithm and perform
 s\nsurprisingly well in a simulated problem of sparse recovery.\n\n\nhttp:
 //www.princeton.edu/~rigollet/index.html
LOCATION:MR12\, CMS\, Wilberforce Road\, Cambridge\, CB3 0WB
END:VEVENT
END:VCALENDAR