Survey Paper 자료정리

1 more property

1. Continual learning 관련 최신 연구 목록화

1.1 Continual learning 벤치마크 데이터셋

CORe50

CORe50는 Continual learning 알고리즘의 object detection 성능을 비교하고 평가하기 위한 표준 벤치마크로 사용된다. CORe50 데이터셋은 50개의 서로 다른 객체 클래스로, 각 클래스는 다양한 각도, 조명 조건 및 배경에서 촬영된 여러 장의 이미지로 구성된다.(Kinect 2.0 센서를 사용하여 300 RGB-D 프레임(20fps)을 수집하여 깊이 정보를 포함한 이미지 데이터를 제공한다.)

지속 학습 시나리오: CORe50는 다양한 Continual learning 시나리오를 제공하여 모델이 새로운 데이터를 점진적으로 학습하면서 이전에 학습한 knowledge를 잊지 않도록 한다.

•

New Instances (NI): 기존 클래스의 새로운 인스턴스를 추가하여 학습.

•

New Classes (NC): 새로운 클래스를 추가하여 학습. 

•

New Instances and Classes (NIC): 새로운 인스턴스와 클래스를 동시에 추가하여 학습.

CORe50

1. Dataset 2. Benchmark 2.1 Differences with other common CL Benchmarks 3. Download 3.1 Dataset 3.2 Object Detection 3.3 Object Segmentation 3.4 Configuration Files 3.5 Results 3.6 Code and Data Loaders 4. Contacts

https://vlomonaco.github.io/core50/index.html#dataset

CLEAR

•

기존 벤치마크 (e.g., ImageNet, MS-COCO)는 정적인 작업을 가정하며, 실제 환경의 동적인 특성을 반영하지 않는다.

•

CLEAR는 2004년부터 2014년까지 10년간의 시각적 개념의 진화를 포함하여, 웹 이미지 컬렉션에서 자연스럽게 발생하는 시간적 변화를 반영한다

The CLEAR Benchmark: Continual LEArning on Real-World Imagery

Please check out our new wikipedia page for full documentation on CLEAR-10 and CLEAR-100 (released in July 2022). It comes with download links to the train set and (newly released) test set of CLEAR-10 and CLEAR-100, as well as summary of the 1st CLEAR challenge hosted on CVPR'22 Open World Vision Workshop and AICrowd. This site is preserved as a presentation of our NeurIPS'21 paper on The CLEAR Benchmark: Continual LEArning on Real-World Imagery.

https://clear-benchmark.github.io/

벤치마크 데이터셋	저자	저널	인용수	년도
CORe50: a New Dataset and Benchmark for Continuous Object Recognition	V Lomonaco, D Maltoni	Conference on robot learning. PMLR	528	2017
The clear benchmark: Continual learning on real-world imagery	Z Lin, J Shi, D Pathak, D Ramanan	NeurIPS Datasets and Benchmarks Track	81	2021
Cglb: Benchmark tasks for continual graph learning	X Zhang, D Song, D Tao	NeurIPS Datasets and Benchmarks Track	36	2022

1.2 최신 Survey 논문

논문	저자	저널	인용수	년도
A comprehensive survey of continual learning: theory, method and application	L Wang, X Zhang, H Su, J Zhu	IEEE Transactions on Pattern Analysis and Machine Intelligence	295	2024
Continual learning: Applications and the road forward	E Verwimp et al.	-	10	2023

1.3 Research

논문	저자	저널	인용수	년도
Embracing change: Continual learning in deep neural networks	R Hadsell, D Rao, AA Rusu, R Pascanu	Trends in cognitive sciences	485	2020
Latent replay for real-time continual learning	L Pellegrini, G Graffieti, et al.	IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)	150	2020
Afec: Active forgetting of negative transfer in continual learning	L Wang, M Zhang, Z Jia, Q Li, C Bao	NeurIPS	64	2021
Towards better plasticity-stability trade-off in incremental learning: A simple linear connector	G Lin, H Chu, H Lai	IEEE/CVF	39	2022
The challenges of continuous self-supervised learning	S Purushwalkam, P Morgado, A Gupta	ECCV	39	2022
Continual learning through retrieval and imagination	Z Wang, L Liu, Y Duan, D Tao	AAAI	34	2022
Maintaining discrimination and fairness in class incremental learning	B Zhao, X Xiao, G Gan, B Zhang	IEEE/CVF	429	2020
Ss-il: Separated softmax for incremental learning	H Ahn, J Kwak, S Lim, H Bang	IEEE/CVF	188	2021
Distilling causal effect of data in class-incremental learning	X Hu, K Tang, C Miao, XS Hua	IEEE/CVF	197	2021
Dark experience for general continual learning: a strong, simple baseline	P Buzzega, M Boschini, A Porrello	NeurIPS	748	2020
Dualnet: Continual learning, fast and slow	Q Pham, C Liu, S Hoi	NeurIPS	149	2021
Co2l: Contrastive continual learning	H Cha, J Lee, J Shin	IEEE/CVF	305	2021
Representational continuity for unsupervised continual learning	D Madaan, J Yoon, Y Li, Y Liu, SJ Hwang	-	-	2021

2. FSCIL 관련 최신 연구 목록화

논문명	저자	저널	인용수	년도
Semantic-aware knowledge distillation for few-shot class-incremental learning	Cheraghian, Ali, et al.	Proceedings of the IEEE/CVF conference on computer vision and pattern recognition	185	2021
Neural collapse inspired feature-classifier alignment for few-shot class incremental learning	Y Yang, H Yuan, X Li, Z Lin, P Torr, D Tao	-	82	2023
Mamba-FSCIL: Dynamic Adaptation with Selective State Space Model for Few-Shot Class-Incremental Learning	X Li, Y Yang, J Wu, B Ghanem, et al.	-	-	2024
Few-shot Tuning of Foundation Models for Class-incremental Learning	S Roy, E Dolatabadi, A Afkanpour, A Etemad	-	-	2024
Forward compatible few-shot class-incremental learning	DW Zhou, FY Wang, HJ Ye, L Ma	IEEE/CVF	174	2022
Overcoming catastrophic forgetting in incremental few-shot learning by finding flat minima	G Shi, J Chen, W Zhang, LM Zhan	NeurIPS	122	2021
Few-shot class-incremental learning via class-aware bilateral distillation	L Zhao, J Lu, Y Xu, Z Cheng, D Guo	IEEE/CVF	30	2023
Constrained few-shot class-incremental learning	M Hersche, G Karunaratne	IEEE/CVF	124	2022
A survey on few-shot class-incremental learning	S Tian, L Li, W Li, H Ran, X Ning, P Tiwari	Elsevier/Neural Networks	32	2024