BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:ML talk of Professor Panagiotis (Panos) Patrinos (KU Leuven)
DTSTART:20190910T110000
DTEND:20190910T120000
DTSTAMP:20260916T034811Z
UID:8b05359bdacae6c9e832443f9f81da24ee7ebdac258e058e7a803c1e
CATEGORIES:Conferences - Seminars
DESCRIPTION:Professor Panagiotis (Panos) Patrinos (KU Leuven)\nTITLE  "Pro
 ximal envelopes for nonconvex splitting algorithms: Block coordinate and N
 ewton-type variants"\n\nABSTRACT The classical understanding of splitting 
 algorithms hinges on monotone operator theory\, and thus heavily hinges on
  convexity. Alternatively\, when applying these schemes to nonconvex probl
 ems convergence can be shown once a suitable merit function is identified\
 ; in other words\, a function that decreases along the generated iterates.
  The challenge of convergence analysis of splitting algorithms in the no
 nconvex setting is thus the identification of a suitable merit function\, 
 and so far there doesn't seem to be a clear way as to how to construct one
 \, let alone for randomized and/or block-coordinate variants. As a result
 \, for many algorithms it is still unclear whether or not their applicatio
 n to nonconvex problems is feasible. In this talk we show how proximal env
 elopes provide a positive answer to this challenge\, as they serve as suit
 able merit functions for many splitting algorithms. Moreover\, thanks to t
 heir regularity properties they enable the possibility to robustify spli
 tting algorithms by means of second-order-type information\, stemming for 
 instance from quasi-Newton schemes\, without affecting global convergence.
  Block-coordinate (BC) variants of forward-backward splitting are also inv
 estigated for the minimization of the sum of a separable smooth function a
 nd a (nonseparable) nonsmooth function\, both of which allowed to be nonco
 nvex. Differently from classical study cases where it is the nonsmooth fun
 ction to be separable\, the cost cannot serve as merit function and as a r
 esult this setting is only little known. Once again we show how the "forwa
 rd-backward envelope" serves as the suitable merit function\, providing ne
 w convergence result for a large class of BC-type algorithms for  nonsmoo
 th and nonconvex problems with rather general sampling strategies\, and t
 hat include the popular Finito/MISO algorithm as a special case.\n\nBIO Pa
 nagiotis (Panos) Patrinos is assistant professor at the Department of Ele
 ctrical Engineering (ESAT) of KU Leuven\, Belgium. In 2014 he was a visi
 ting professor at Stanford University. He received his PhD in Control and
  Optimization\, M.S. in Applied Mathematics and M.Eng. from the National 
 Technical University of Athens in 2010\, 2005 and 2003\, respectively. Af
 ter his PhD he held postdoc positions at the University of Trento and IMT 
 Lucca\, Italy\, where he became an assistant professor in 2012. His curren
 t research interests lie in the intersection of optimization control and 
 learning. In particular he is interested in the theory and algorithms for 
 structured convex and nonconvex optimization as well as learning-based\, r
 isk-averse model predictive control with a wide range of applications inc
 luding autonomous vehicles\, smart grids\, water networks\, aerospace\, m
 ulti-agent systems\, signal processing and machine learning.
LOCATION:ELD 120 https://plan.epfl.ch/?room==ELD%20120
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR