BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//EVTZ//Calendar Feed//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
X-WR-CALNAME:Admin 1 - EVTZ export
BEGIN:VEVENT
UID:event-209607@evtz.net
X-EVTZ-EVENT-ID:209607
DTSTAMP:20261110T031913Z
DTSTART:20261112T113000Z
DTEND:20261112T123000Z
SUMMARY:Why Adam work in optimizing language models: heavy-tailed data imba
 lance
DESCRIPTION:Adam and related methods are the default optimizers in training
  language models because gradient descent is so slow it is thought it simpl
 y does not work. Many methods have been proposed to improve on Adam\, but w
 e have a poor understanding of why it works better in the first plac\n\nAda
 m and related methods are the default optimizers in training language model
 s because gradient descent is so slow it is thought it simply does not work
 . Many methods have been proposed to improve on Adam\, but we have a poor u
 nderstanding of why it works better in the first place. This talk will be a
 bout understanding what makes language models hard for gradient descent\, a
 nd how Adam fixes it. We\n\nPrice / conditions: free\n\nResponsible: ETH AI
  Center\n\nSource: hackevents.net Switzerland API\n\nSource URL: https://ha
 ckevents.net/api/events?country=ch&limit=50
LOCATION:OAT\, Room 11\, ETH AI Center Foyer\, Zurich
URL:https://evtz.net/events/209607
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR