Transferable Attention for Domain Adaptation

By Ximei Wang, Liang Li, Weirui Ye, Mingsheng Long, Jianmin Wang. AAAI Conference on Artificial Intelligence, 2019

314 citations recorded by the source corpus.

Recent work in domain adaptation bridges different domains by adversarially learning a domain-invariant representation that cannot be distinguished by a domain discriminator. Existing methods of adversarial domain adaptation mainly align the global images across the source and target domains. However, it is obvious that not all regions of an image are transferable, while forcefully aligning the untransferable regions may lead to negative transfer. Furthermore, some of the images are significantly dissimilar across domains, resulting in weak image-level transferability. To this end, we present Transferable Attention for Domain Adaptation (TADA), focusing our adaptation model on transferable regions or images. We implement two types of complementary transferable attention: transferable local attention generated by multiple region-level domain discriminators to highlight transferable regions, and transferable global attention generated by single image-level domain discriminator to highlight transferable images. Extensive experiments validate that our proposed models exceed state of the art results on standard domain adaptation datasets.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Radiology, Nuclear Medicine and Imaging

View Paper: https://www.semanticscholar.org/paper/fa2c5d0ff15655dfb111c6d0e0bfc4116d003851

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Luc Van Gool
2. Mingkui Tan
3. Diane Joyce Cook
4. Garrett Wilson
5. Mingsheng Long
6. Jianmin Wang
7. Zhengming Ding
8. K. Jia

In Computer Vision and Pattern Recognition

1. Luc Van Gool
2. Ming-Hsuan Yang
3. Dacheng Tao
4. Yi Yang
5. Ling Shao
6. Xiaogang Wang
7. Trevor Darrell
8. Jiashi Feng

In Radiology, Nuclear Medicine and Imaging

1. Ying Jin
2. Ximei Wang
3. K. Jia
4. Hui Tang
5. Mingsheng Long
6. Jianmin Wang
7. Qi-Yan Dou
8. Diane Joyce Cook

Citations in the corpus

• Collaborative Unsupervised Domain Adaptation for Medical Image Diagnosis (2019, methodology): “Existing UDA methods [27]–[29] for deep learning reduce the domain”
• Mining Cross-Image Semantics for Weakly Supervised Semantic Segmentation (2020, background): “With their popularity in the field of natural language processing [7,37,41,47,59], attention modeling is rapidly adopted in various computer vision tasks, such as image recognition [12,21,57,64,73], domain adaptation [65,84], human pose estimation [8, 61, 78], reasoning among objects [52, 87], and i”
• Unsupervised Domain Adaptation via Structurally Regularized Deep Clustering (2020, background): “3 TADA [56] 94.”
• HGNet: Hybrid Generative Network for Zero-Shot Domain Adaptation (2020, background): “However, domain adaptation provides an alternative strategy to handle with such a problem [24, 9, 28].”
• HoMM: Higher-order Moment Matching for Unsupervised Domain Adaptation (2019, result): “Note that the experimental results of the compared methods are reported from (Wang et al. 2019) directly.”
• Temporal Attentive Alignment for Large-Scale Video Domain Adaptation (2019, methodology): “Our proposed TA(3)N and TADA [50] both utilize entropy functions for attention but with different perspectives.”
• Harmonizing Transferability and Discriminability for Adapting Object Detectors (2020, methodology): “With insights from the practice of Generative Adversarial Nets (GAN) [16], tremendous works [14, 54, 38, 45, 56, 5] have been done by leveraging the two-player game to achieve domain confusion with Gradient Reversal Layer (GRL) for feature alignment.”
• A Survey of Unsupervised Deep Domain Adaptation (2018, background): “[247] explore the use of attention networks, Kang et al.”
• Minimum Class Confusion for Versatile Domain Adaptation (2019, methodology): “TADA [47] introduces the first transferable attention mechanism.”
• Catastrophic Forgetting Meets Negative Transfer: Batch Spectral Shrinkage for Safe Transfer Learning (2019, background): “However, negative transfer, a major challenge in domain adaptation [31, 34, 38, 40, 41], has rarely been considered in inductive transfer learning.”
• Spatial Attention Pyramid Network for Unsupervised Domain Adaptation (2020, background): “To highlight transferable regions in domain adaptation, Wang et al. [30] use multiple region-level domain discriminators and single image-level domain discriminator to generate transferable local and global attention, respectively.”
• Discriminative Partial Domain Adversarial Network (2020, background): “The basic assumption of standard DA is that source and target domain share the same label space [19, 30], which often does not hold in practice.”