BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:talk of Professor Marco Mondelli (IST Austria)
DTSTART:20220421T161500
DTEND:20220421T171500
DTSTAMP:20260930T215105Z
UID:b8bfb55a8c4d7c52c7aa9364e2a4645fbe35c9c80f769041bdf03aa3
CATEGORIES:Conferences - Seminars
DESCRIPTION:Professor Marco Mondelli\nTitle:\nUnderstanding Gradient Desce
 nt for Over-parameterized Deep Neural Networks\n\nAbstract:\nTraining a ne
 ural network is a non-convex problem that exhibits spurious and disconnect
 ed local minima. Yet\, in practice neural networks with millions of parame
 ters are successfully optimized using gradient descent methods. In this ta
 lk\, I will give some theoretical insights on why this is possible and dis
 cuss two approaches to study the behavior of gradient descent. The first o
 ne takes a mean-field view and it relates the dynamics of stochastic gradi
 ent descent (SGD) to a certain Wasserstein gradient flow in probability sp
 ace. I will show how this idea allows to study the connectivity\, converge
 nce and implicit bias of the solutions found by SGD. The second approach c
 onsists in the analysis of the Neural Tangent Kernel. I will present tight
  bounds on its smallest eigenvalue and show their implications on memoriza
 tion and optimization in deep networks.\n\nBased on joint work with Adel J
 avanmard\, Vyacheslav Kungurtsev\, Andrea Montanari\, Guido Montufar\, Quy
 nh Nguyen\, and Alexander Shevchenko.\n\nBio:\nMarco Mondelli received the
  B.S. and M.S. degree in Telecommunications Engineering from the Universit
 y of Pisa\, Italy\, in 2010 and 2012\, respectively. In 2016\, he obtained
  his Ph.D. degree in Computer and Communication Sciences at the École Pol
 ytechnique Fédérale de Lausanne (EPFL)\, Switzerland. He is currently an
  Assistant Professor at the Institute of Science and Technology Austria (I
 ST Austria). Prior to that\, he was a Postdoctoral Scholar in the Departme
 nt of Electrical Engineering at Stanford University\, USA\, from February 
 2017 to August 2019. He was also a Research Fellow with the Simons Institu
 te for the Theory of Computing\, UC Berkeley\, USA\, for the program on Fo
 undations of Data Science from August to December 2018. His research inter
 ests include data science\, machine learning\, information theory\, wirele
 ss communication systems\, and modern coding theory. He was the recipient 
 of a number of fellowships and awards\, including the Jack K. Wolf ISIT St
 udent Paper Award in 2015\, the STOC Best Paper Award in 2016\, the EPFL D
 octorate Award in 2018\, the Simons-Berkeley Research Fellowship in 2018\,
  the Lopez-Loreta Prize in 2019\, and Information Theory Society Best Pape
 r Award in 2021.
LOCATION:INR 219 https://plan.epfl.ch/?room==INR%20219
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR