BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:EE Seminar: Prospects and Challenges for Machine Learning in the P
 hysical World
DTSTART:20210430T150000
DTEND:20210430T160000
DTSTAMP:20260408T060340Z
UID:0eb1d396932e4e0b73223b8b07adadb4823b78e27e99fefc5907c7ff
CATEGORIES:Conferences - Seminars
DESCRIPTION:Joan Bruna is an Associate Professor of Computer Science\, Dat
 a Science and Mathematics (affiliated) at the Courant Institute and the C
 enter for Data Science\, New York University (NYU)\, and a visiting schola
 r at the Flatiron Institute. Previously\, he was Assistant Professor of S
 tatistics at UC Berkeley and part of BAIR (Berkeley AI Research). He comp
 leted his PhD in 2013 at Ecole Polytechnique\, France. Before his PhD he 
 was a Research Engineer at a semi-conductor company\, developing real-time
  video processing algorithms. Even before that\, he did a MsC at Ecole No
 rmale Superieure de Cachan in Applied Mathematics (MVA) and a BA and MS a
 t UPC (Universitat Politecnica de Catalunya\, Barcelona) in both Mathemati
 cs and Telecommunication Engineering. For his research contributions\, he
  has been awarded a Sloan Research Fellowship (2018)\, a NSF CAREER Award
  (2019)\, a best paper award at ICMLA (2018) and the IAA Outstanding Pape
 r Award.\nAbstract: The last decade has witnessed an experimental revoluti
 on in data science\, led by the huge empirical success of deep learning m
 ethods across many areas of science and engineering. In order to capitali
 se on these successes\, it has become increasingly important to provide a 
 mathematical foundation that gives guiding design principles\, and mitiga
 tes the current data ‘hunger’ of these DL architectures\, to enable f
 urther applications within computational science. \n\nIn this talk\, I wi
 ll describe the crucial role that data structure plays in constructing suc
 h foundations. Existing mathematical models are mostly agnostic to data s
 tructure\, and as a result rely on strong assumptions in order to break t
 he curse of dimensionality. Alternatively\, I will present a geometrical p
 erspective that unifies all successful DL architectures (CNNs\, RNNs\, Tr
 ansformers\, GNNs) from the principles of symmetry and scale separation\,
  providing a viable mathematical picture where the curse of dimensionality
  is avoided under more realistic assumptions. I will cover applications o
 f such principled geometric models in graph inference\, 3d surface recons
 truction\, cosmology and quantum mechanics\, and describe current mathemat
 ical challenges towards further integrating DL within the computational s
 ciences.
LOCATION:https://epfl.zoom.us/j/87406224364
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR