Why Adam work in optimizing language models: heavy-tailed data imbalance
Adam and related methods are the default optimizers in training language models 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 understanding of why it works better in the first plac
This event is also part of the public calendar streams below, so visitors can return to the broader series instead of only this single page.
Time
12 Nov 2026 11:30 to 12 Nov 2026 12:30
Venue
OAT, Room 11, ETH AI Center Foyer, Zurich
OAT, Room 11, ETH AI Center Foyer, Zurich, ch
Publisher
ETH AI Center
Public calendars carrying this event
Duration
1 hour
Responsible
ETH AI Center
Source details
- Import status
- Regional Scan
- Source kind
- Native
- Updated
- 2 hours ago
- Public calendars
- Switzerland Tech Events
Event map
OAT, Room 11, ETH AI Center Foyer, Zurich, ch
Map data © OpenStreetMap contributors
Adam and related methods are the default optimizers in training language models 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 understanding of why it works better in the first place. This talk will be about understanding what makes language models hard for gradient descent, and how Adam fixes it. We Price / conditions: free