BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:IC Colloquium : Approximation Algorithms for Large-Scale Data Anal
 ysis
DTSTART:20170327T101500
DTEND:20170327T113000
DTSTAMP:20260407T020807Z
UID:ccaf924d1640b42bd3d76f630b7df614dbe9115f6d75a65d7489c6f4
CATEGORIES:Conferences - Seminars
DESCRIPTION:By : Ludwig Schmidt - MIT\nIC Faculty candidate\n\nAbstract :\
 nOver the past decade\, the scale of computational problems in machine lea
 rning has grown tremendously. At the same time\, the slowdown of Moore’s
  Law is making it harder to apply existing algorithms to large data sets. 
 As this trend will only increase in the foreseeable future\, we need faste
 r algorithms for many computational problems in machine learning.\n \nIn 
 this talk\, I will show how ideas from approximation algorithms can be use
 d to speed up multiple key tasks in machine learning. Since the underlying
  statistical problems are inherently noisy\, computing a good approximate 
 solution often leads to significantly faster algorithms that match the sta
 tistical performance of their exact counterparts. One such connection is i
 n the area of constrained estimation\, which underlies many problems such 
 as compressive sensing\, sparse linear regression\, and matrix completion.
  By introducing approximate projections\, I will connect classical tools f
 rom theoretical computer science with these estimation tasks.\n \nIn the 
 second part of the talk\, I will turn to classification and give strong ev
 idence that approximation is inherently necessary in order to get sub-quad
 ratic running time for widely used learning algorithms such as kernel meth
 ods and neural networks. On the other hand\, I will show how ideas from ne
 arest neighbor algorithms can exploit problem structure and speed up large
 -multiclass networks.\n\nBio :\nLudwig Schmidt is a PhD student at MIT\, a
 dvised by Prof. Piotr Indyk. Ludwig’s research interests revolve around 
 algorithmic aspects of machine learning\, statistics\, and signal processi
 ng. Ludwig received a Google PhD Fellowship in machine learning\, a Simons
 -Berkeley research fellowship\, and a best paper award at the Internationa
 l Conference on Machine Learning (ICML).\n\nMore information\n\n\n 
LOCATION:BC 420 https://plan.epfl.ch/?room==BC%20420
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR