Memory-Efficient Continual Learning with CLIP Models

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: King, Ryan, Li, Gang, Mortazavi, Bobak, Yang, Tianbao
Format: Preprint
Publié: 2026
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866911649107017728
author King, Ryan
Li, Gang
Mortazavi, Bobak
Yang, Tianbao
author_facet King, Ryan
Li, Gang
Mortazavi, Bobak
Yang, Tianbao
contents Contrastive Language-Image Pretraining (CLIP) models excel at understanding image-text relationships but struggle with adapting to new data without forgetting prior knowledge. To address this, models are typically fine-tuned using both new task data and a memory buffer of past tasks. However, CLIP's contrastive loss suffers when the memory buffer is small, leading to performance degradation on previous tasks. We propose a memory-efficient, distributionally robust method that dynamically reweights losses per class during training. Our approach, tested on class incremental settings (CIFAR-100, ImageNet1K) and a domain incremental setting (DomainNet) adapts CLIP models quickly while minimizing catastrophic forgetting, even with minimal memory usage.
format Preprint
id arxiv_https___arxiv_org_abs_2605_03866
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Memory-Efficient Continual Learning with CLIP Models
King, Ryan
Li, Gang
Mortazavi, Bobak
Yang, Tianbao
Machine Learning
Contrastive Language-Image Pretraining (CLIP) models excel at understanding image-text relationships but struggle with adapting to new data without forgetting prior knowledge. To address this, models are typically fine-tuned using both new task data and a memory buffer of past tasks. However, CLIP's contrastive loss suffers when the memory buffer is small, leading to performance degradation on previous tasks. We propose a memory-efficient, distributionally robust method that dynamically reweights losses per class during training. Our approach, tested on class incremental settings (CIFAR-100, ImageNet1K) and a domain incremental setting (DomainNet) adapts CLIP models quickly while minimizing catastrophic forgetting, even with minimal memory usage.
title Memory-Efficient Continual Learning with CLIP Models
topic Machine Learning
url https://arxiv.org/abs/2605.03866