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
Questo evento fa parte anche dei flussi pubblici di calendario qui sotto, così i visitatori possono tornare alla serie completa.
Orario
12 Nov 2026 11:30 a 12 Nov 2026 12:30
Luogo
OAT, Room 11, ETH AI Center Foyer, Zurich
OAT, Room 11, ETH AI Center Foyer, Zurich, ch
Pubblicatore
ETH AI Center
Calendari pubblici che includono questo evento
Duration
1 ora
Responsible
ETH AI Center
Source details
- Import status
- Regional Scan
- Source kind
- Nativo
- Updated
- 1 ora fa
- Public calendars
- Switzerland Tech Events
Mappa evento
OAT, Room 11, ETH AI Center Foyer, Zurich, ch
Dati mappa © collaboratori OpenStreetMap
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