BEGIN:VCALENDAR
VERSION:2.0
PRODID:icalendar-ruby
CALSCALE:GREGORIAN
X-WR-CALNAME:Stochastic Optimization via Online Learning
X-WR-TIMEZONE:Eastern Time (US & Canada)
BEGIN:VEVENT
DTSTAMP:20260912T225420Z
UID:tag:localist.com\,2008:EventInstance_45925218937024
DTSTART:20240328T150000Z
DTEND:20240328T160000Z
DESCRIPTION:Stochastic optimization algorithms like gradient descent are a 
 fundamental tool in modern machine learning. However\, our understanding o
 f why some algorithms are better than others is woefully incomplete - in m
 any cases the empirical evidence seems to directly contradict standard the
 oretical guidelines! In this talk\, I will highlight some fundamental gaps
  in our knowledge\, and show how we can partially address these gaps using
  tools from online learning. Online learning is a seemingly unrelated fiel
 d that studies sequential decision making in a potentially adversarial env
 ironment. Perhaps surprisingly\, online learning is connected to many area
 s of machine learning\, including stochastic optimization. However\, these
  classical results do not shed too much light on modern empirical techniqu
 es. By building new and improved reductions from stochastic optimization t
 o online learning\, we will gain a better understanding of current algorit
 hms used in practice\, and even find directions for future algorithm desig
 n.\n\nLocation: SEC 1.307 and Zoom (Password: 599436)
GEO:42.363197;-71.127278
LOCATION:Science and Engineering Complex (SEC)\, SEC 1.307
SUMMARY:Stochastic Optimization via Online Learning
URL;VALUE=URI:https://events.seas.harvard.edu/event/stochastic-optimization
 -via-online-learning
CATEGORIES:Colloquia / Seminar / Lecture
END:VEVENT
END:VCALENDAR