논문 리뷰
1.[CVPR 2023] SegGPT 논문 리뷰
PaperGithub Segment everything with a Generalist Painter 의 약자로 Painter 라는 아키텍쳐에 기반으로 image의 semantic 정보를 학습해서 segmentation 을 해주는 모델.→ GPT(Generative p
2.LoRA(Low-Rank Adaptation)
over-parameterized model이 실제로 low intrinsic dimension에 있다는 논문들에 영향을 받음low intrinsic dimension = 과연 파라미터 수에 대해 밥값을 하고 있느냐라고 생각하면 됨LoRA를 사용하면Pre-traine
3.Lifelong Learning with Dynamically Expandable Networks
이 논문은 LIFELONG LEARNING을 위한 Dynamically Expandable Network(DEN)을 제안한다.위 Network는 training에 있어 Capacity를 Dynamically하게 결정하여 Compact한 overlapping된 지식을 공
4.LIME(Local Interpretable Model-agnostic Explanations) BackGround
논문 링크: https://arxiv.org/abs/1602.04938 LIME은 Local | Pertubation | Post-Hoc(agnostic)의 종류에 속한다.Post-Hoc은 Surrogate Model을 사용하는데 이는 Black box mod
5.SmoothGrad: removing noise by adding noise
링크 : https://github.com/lbg030/velog/tree/main/AI-SmoothGrad\_%20removing%20noise%20by%20adding%20noise-081223-015826
6.Grad CAM
Github: https://github.com/lbg030/velog
7.Basic of Active Learning
https://github.com/lbg030/velog
8.Expected Model Change
https://github.com/lbg030/velog
9.Core-Set Algorithm
https://github.com/lbg030/velog
10.Learning Loss for AL
https://github.com/lbg030/velog
11.PT4AL
https://github.com/lbg030/velog
12.Query By Committee(QBC)
https://github.com/lbg030/velog