MgSvF: Multi-Grained Slow vs. Fast Framework for Few-Shot Class-Incremental Learning

Zhao, Huimin; Fu, Yongjian; Mintong, Kang,; Wu, Fei; Li, Xi

doi:10.48550/arxiv.2006.15524

Cited by 1 publication

(2 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Incremental few-shot learning [47,41,60,9,8] aims to incrementally learn from very few samples. TOPCI [47] proposes a neural gas network to learn and preserve the topology of the feature manifold formed by different classes.…”

Section: Related Workmentioning

confidence: 99%

“…Current research. The study of incremental few-shot learning has just started [47,41,60,9,8,34,59]. Current works mainly borrow ideas from research in incremental learning to overcome the forgetting problem, by enforcing strong constraints on model parameters to penalize the changes of parameters [34,28,56], or by saving a small amount of exemplars from old classes and adding constraints on the exemplars to avoid forgetting [40,20,4].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Overcoming Catastrophic Forgetting in Incremental Few-Shot Learning by Finding Flat Minima

Shi¹,

Chen²,

Zhang³

et al. 2021

Preprint

View full text Add to dashboard Cite

This paper considers incremental few-shot learning, which requires a model to continually recognize new categories with only a few examples provided. Our study shows that existing methods severely suffer from catastrophic forgetting, a well-known problem in incremental learning, which is aggravated due to data scarcity and imbalance in the few-shot setting. Our analysis further suggests that to prevent catastrophic forgetting, actions need to be taken in the primitive stagethe training of base classes instead of later few-shot learning sessions. Therefore, we propose to search for flat local minima of the base training objective function and then fine-tune the model parameters within the flat region on new tasks. In this way, the model can efficiently learn new classes while preserving the old ones. Comprehensive experimental results demonstrate that our approach outperforms all prior state-of-the-art methods and is very close to the approximate upper bound. The source code is available at https://github.com/moukamisama/F2M.

show abstract

Section: Related Workmentioning

confidence: 99%