BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//132.216.98.100//NONSGML kigkonsult.se iCalcreator 2.20.4//
BEGIN:VEVENT
UID:20260722T231138EDT-2649U6i9DI@132.216.98.100
DTSTAMP:20260723T031138Z
DESCRIPTION:Virtual Informal Systems Seminar (VISS)\n	Centre for Intelligent
  Machines (CIM) and Groupe d'Etudes et de Recherche en Analyse des Decisio
 ns (GERAD)\n	\n	Zoom Link\n	Meeting ID: 910 7928 6959        \n	Passcode: VISS
   \n	\n	Speaker: Petar Veličković\, Staff Research Scientist\, DeepMind\, Lo
 ndon\, UK\n	\n	Abstract: \n	The last decade has witnessed an experimental rev
 olution in data science and machine learning\, epitomised by deep learning
  methods. Indeed\, many high-dimensional learning tasks previously thought
  to be beyond reach –such as computer vision\, playing Go\, or protein fol
 ding – are in fact feasible with appropriate computational scale. Remarkab
 ly\, the essence of deep learning is built from two simple algorithmic pri
 nciples: first\, the notion of representation or feature learning\, whereb
 y adapted\, often hierarchical\, features capture the appropriate notion o
 f regularity for each task\, and second\, learning by local gradient-desce
 nt type methods\, typically implemented as backpropagation.\n	\n	While learn
 ing generic functions in high dimensions is a cursed estimation problem\, 
 most tasks of interest are not generic\, and come with essential pre-defin
 ed regularities arising from the underlying low-dimensionality and structu
 re of the physical world. This talk is concerned with exposing these regul
 arities through unified geometric principles that can be applied throughou
 t a wide spectrum of applications.\n	\n	Such a 'geometric unification' endea
 vour in the spirit of Felix Klein's Erlangen Program serves a dual purpose
 : on one hand\, it provides a common mathematical framework to study the m
 ost successful neural network architectures\, such as CNNs\, RNNs\, GNNs\,
  and Transformers. On the other hand\, it gives a constructive procedure t
 o incorporate prior physical knowledge into neural architectures and provi
 de principled way to build future architectures yet to be invented.\n	\n	Bio
 graphy: \n\nDr. Veličković is a Staff Research Scientist at DeepMind\, Aff
 iliated Lecturer at the University of Cambridge\, and an Associate of Clar
 e Hall\, Cambridge. He holds a PhD in Computer Science from the University
  of Cambridge (Trinity College)\, obtained under the supervision of Pietro
  Liò. His research concerns geometric deep learning—devising neural networ
 k architectures that respect the invariances and symmetries in data. For h
 is contributions\, he is recognised as an ELLIS Scholar in the Geometric D
 eep Learning Program. Particularly\, he focuses on graph representation le
 arning and its applications in algorithmic reasoning (featured in VentureB
 eat). He is the first author of Graph Attention Networks—a popular convolu
 tional layer for graphs—and Deep Graph Infomax—a popular self-supervised l
 earning pipeline for graphs (featured in ZDNet). His research has been use
 d in substantially improving travel-time predictions in Google Maps (featu
 red in the CNBC\, Endgadget\, VentureBeat\, CNET\, the Verge and ZDNet)\, 
 and guiding intuition of mathematicians towards new top-tier theorems and 
 conjectures (featured in Nature\, Science\, Quanta Magazine\, New Scientis
 t\, The Independent\, Sky News\, The Sunday Times\, la Repubblica and The 
 Conversation). \n\nHomepage\n
DTSTART:20220916T180000Z
DTEND:20220916T190000Z
LOCATION:CA\, ZOOM
SUMMARY:Geometric Deep Learning
URL:https://www.mcgill.ca/cim/channels/event/geometric-deep-learning-340877
END:VEVENT
END:VCALENDAR