Domain Adaptation
Domain adaptation addresses the challenge of applying machine learning models trained on one dataset (the source domain) to a different dataset with a different distribution (the target domain). Current research focuses on techniques like adversarial training, knowledge distillation, and optimal transport to bridge this domain gap, often employing transformer-based models, generative adversarial networks (GANs), and various meta-learning approaches. This field is crucial for improving the robustness and generalizability of machine learning models across diverse real-world applications, particularly in areas with limited labeled data such as medical imaging, natural language processing for low-resource languages, and personalized recommendation systems. The development of standardized evaluation frameworks is also a growing area of focus to ensure fair comparison and reproducibility of results.
Papers
A Client-server Deep Federated Learning for Cross-domain Surgical Image Segmentation
Ronast Subedi, Rebati Raman Gaire, Sharib Ali, Anh Nguyen, Danail Stoyanov, Binod Bhattarai
SMC-UDA: Structure-Modal Constraint for Unsupervised Cross-Domain Renal Segmentation
Zhusi Zhong, Jie Li, Lulu Bi, Li Yang, Ihab Kamel, Rama Chellappa, Xinbo Gao, Harrison Bai, Zhicheng Jiao
Cross-Database and Cross-Channel ECG Arrhythmia Heartbeat Classification Based on Unsupervised Domain Adaptation
Md Niaz Imtiaz, Naimul Khan
SF-FSDA: Source-Free Few-Shot Domain Adaptive Object Detection with Efficient Labeled Data Factory
Han Sun, Rui Gong, Konrad Schindler, Luc Van Gool
Text-only Domain Adaptation using Unified Speech-Text Representation in Transducer
Lu Huang, Boyu Li, Jun Zhang, Lu Lu, Zejun Ma