方法对比

	域自适应 Transformer	Vision Transformer
领域	深度学习	深度学习
方法族	Machine learning	Machine learning
起源年份	2019–2022	2021
提出者	Various (Vaswani et al. 2017 for Transformers; domain adaptation extensions emerged 2019–2022)	Dosovitskiy, A. et al.
类型	Pre-trained model fine-tuned with domain-shift adaptation	Transformer architecture for images (self-attention over patches)
开创性文献	Ni, J., Hernandez Abrego, G., Constant, N., Ma, J., Hall, K., Cer, D., & Yang, Y. (2021). Sentence-T5: Scalable Sentence Encoders from Pre-trained Text-to-Text Models. Findings of ACL 2022. arXiv:2108.08877.	Dosovitskiy, A. et al. (2021). An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale. ICLR.
别名	DAT, domain-adaptive Transformer, domain adaptation with Transformers, transfer-learning Transformer	Görsel Transformer (ViT), görsel transformer, ViT, patch transformer for images
相关	2	5
摘要	A Domain-Adaptive Transformer (DAT) is a Transformer-based model — such as BERT or ViT — extended with an explicit domain-alignment objective so that learned representations transfer well from a labeled source domain to a different, often unlabeled, target domain. The approach combines the powerful representation capacity of Transformers with domain adaptation techniques such as adversarial training or contrastive alignment to minimise domain shift.	The Vision Transformer (ViT), introduced by Dosovitskiy and colleagues in 2021, splits an image into fixed-size patches, treats those patches as a sequence, and applies the Transformer self-attention mechanism to image classification. Given enough training data, it surpasses convolutional neural networks (CNNs).