Paper/Continual Learning (2) 썸네일형 리스트형 [CL] End-to-End Incremental Learning 0. Abstract catastrophic forgetting을 #incremental learning 을 통해 해결하고자 함. 전체 frame work 를 E2E으로 구성. 즉, data representation과 classifier를 jointly learn CIFAR-100 , ImageNet을 통해 Evaluate 함. 1. Introduction Main challenge : 현실에 적용하여 incremental하게 학습할 수 있는 classifier를 위한 / visual recognition system 구축 기존의 모델은 new data + old data 조금으로는 학습 불가능함. (시도는 있었으나 성능저하 극심했음.) Incremental DL approach flow of data.. [CL] Lifelong Learning with Dynamically Expandable Networks (ICLR 2018) https://openreview.net/pdf?id=Sk7KsfW0- 0. Abstarct Dynamically Expandable Network (DEN) dynamically decide its network capacity as it trains on a sequence of tasks trained in an online manner by performing selective retraining 1 Introduction lifelong learning → trained in an online manner by performing selective retraining [ Strategy 1 ] Fine-tune : training both origin & new task → degenerate .. 이전 1 다음
