BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:IC Colloquium: The Adaptive Complexity of Submodular Optimization:
  Exponentially Faster Algorithms for Machine Learning and Beyond
DTSTART:20190228T101500
DTEND:20190228T111500
DTSTAMP:20260415T093223Z
UID:07fa1425159518a3ab1278ad4f3dde9196e9a6ffc66017a94d86fcfe
CATEGORIES:Conferences - Seminars
DESCRIPTION:By: Eric Balkanski - Harvard University\nIC Faculty candidate\
 n\nAbstract:\nMany methods in optimization are inherently sequential and c
 onsequently cannot be efficiently parallelized. In this talk\, I will desc
 ribe a novel approach called adaptive sampling that yields algorithms whos
 e parallel running time is exponentially faster than any previous algorith
 m for submodular optimization.  Maximizing a submodular function is the a
 lgorithmic engine behind a growing number of machine learning applications
  such as speech and document summarization\, recommendation systems\, clus
 tering\, feature selection\, and network analysis. \n \nThe speedups are
  in the adaptive complexity model\, which is an information theoretic abst
 raction for parallelism. The concept of adaptivity plays an important role
  in a broader agenda of understanding the limitations and possibilities of
  data-driven decision making. In this talk I will first introduce adaptivi
 ty\, then describe the adaptive sampling algorithms\, and finally present 
 experimental results from various application domains.\n\nBio:\nEric is a 
 fifth year PhD student in Computer Science at Harvard University where he 
 is advised by Yaron Singer. Before his PhD\, he received his B.S. from Car
 negie Mellon University. He is the recipient of a Google PhD fellowship\, 
 a Smith Family Graduate Science and Engineering Fellowship\, a Best Paper 
 Award at the 18th Conference on Implementation and Application of Automata
  in 2013\, and is also an Andrew Carnegie Society Scholar.  His research 
 interests include machine learning\, optimization\, algorithms\, networks\
 , and mechanism design.\n\nMore information
LOCATION:BC 420 https://plan.epfl.ch/?room==BC%20420
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR