Benutzer: Gast  Login
Titel:

Provable Continual Learning via Sketched Jacobian Approximations

Dokumenttyp:
Konferenzbeitrag
Autor(en):
Heckel, Reinhard
Abstract:
An important problem in machine learning is the ability to learn tasks in a sequential manner. If trained with standard first-order methods most models forget previously learned tasks when trained on a new task, which is often referred to as catastrophic forgetting. A popular approach to overcome forgetting is to regularize the loss function by penalizing models that perform poorly on previous tasks. For example, elastic weight consolidation (EWC) regularizes with a quadratic form involving a di...     »
Kongress- / Buchtitel:
International Conference on Artificial Intelligence and Statistics (AISTATS)
Jahr:
2022
WWW:
https://arxiv.org/abs/2112.05095
 BibTeX