BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:IC Colloquium: Rethinking the role of optimization in learning
DTSTART:20190320T101500
DTEND:20190320T111500
DTSTAMP:20260506T061805Z
UID:a02075abf0faf81e0b01e647e17669013363f33dcf77f974e39509f1
CATEGORIES:Conferences - Seminars
DESCRIPTION:By: Suriya Gunsekar - Toyota Technological Institute at Chicag
 o\nIC Faculty candidate\n\nAbstract:\nIn this talk\, I will overview our r
 ecent progress towards understanding how we learn large capacity machine l
 earning models. In the modern practice of machine learning\, especially de
 ep learning\, many successful models have far more trainable parameters co
 mpared to the number of training examples. Consequently\, the optimization
  objective for training such models have multiple minimizers that perfectl
 y fit the training data. More problematically\, while some of these minimi
 zers generalize well to new examples\, most minimizers will simply overfit
  or memorize the training data and will perform poorly on new examples. In
  practice though\, when such ill-posed objectives are minimized using loca
 l search algorithms like (stochastic) gradient descent ((S)GD)\, the "spec
 ial" minimizers returned by these algorithms have remarkably good performa
 nce on new examples. In this talk\, we will explore the role optimization 
 algorithms like (S)GD in learning overparameterized models in simpler sett
 ing of learning linear predictors.\n\nBio:\nSuriya Gunasekar is a research
  assistant professor at the Toyota Technological Institute at Chicago. Pri
 or to joining TTIC\, she finished her PhD at the University of Texas at Au
 stin advised by Prof. Joydeep Ghosh. Her research interests are broadly dr
 iven by statistical\, algorithmic\, and societal aspects of machine learni
 ng including topics of optimization\, high dimensional learning\, and algo
 rithmic fairness. \n\nMore information
LOCATION:BC 420 https://plan.epfl.ch/?room==BC%20420
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR