로딩
요청 처리 중입니다...

NEURAL COLLAPSE INSPIRED FEATURE-CLASSIFIER ALIGNMENT FOR FEW-SHOT CLASS INCREMENTAL LEARNING

 NEURAL COLLAPSE INSPIRED FEATURE-CLASSIFIER ALIGNMENT FOR FEW-SHOT CLASS INCREMENTAL LEARNING

Authors : Yibo Yang1∗†, Haobo Yuan2∗ , Xiangtai Li3 , Zhouchen Lin3,4,5† , Philip Torr6 , Dacheng Tao1 Conf. : ICLR 2022 paper : NEURAL COLLAPSE INSPIRED FEATURE-CLASSIFIER ALIGNMENT FOR FEW-SHOT CLASS INCREMENTAL LEARNING [https://arxiv.org/pdf/2302.03004] NC-FSCIL 정리 Introduction. NC-FSCIL (Neural Collapse-inspired Few-Shot Class-Incremental Learning) 주요 내용: FSCIL (Few-Shot Class-Incremental Learning)의 정의: 새로운 클래스를 점진적으로 학습하면서 각 세션마다 소수의 샘플만 사용 기존 클래스에 대한 성능을 유지하면서 새로운 클래스 학습 기존 클래스를 학습할 때에는 주...

# 2022 # fewshot # FSCIL # ICLR # ICLR2022 # IL # incremental # NCFSCIL # NeuralCollapse # Prototype # FeatureClassifier # ETRPrototype # Alignment # CL # Classifier # continual # continuallearning # DotRegressionLoss # DRloss # Equiangular # ETF # 등각프레임