Domain Shift
Domain shift, the discrepancy between training and deployment data distributions, significantly degrades machine learning model performance. Current research focuses on developing robust algorithms and model architectures, such as U-Nets, Swin Transformers, and diffusion models, to mitigate this issue through techniques like distribution alignment, adversarial training, and knowledge distillation. These efforts are crucial for improving the reliability and generalizability of machine learning models across diverse real-world applications, particularly in medical imaging, autonomous driving, and natural language processing, where data heterogeneity is common. The ultimate goal is to create models that generalize effectively to unseen data, reducing the need for extensive retraining and improving the practical impact of AI systems.
Papers
Gradual Domain Adaptation via Manifold-Constrained Distributionally Robust Optimization
Amir Hossein Saberi, Amir Najafi, Ala Emrani, Amin Behjati, Yasaman Zolfimoselo, Mahdi Shadrooy, Abolfazl Motahari, Babak H. Khalaj
SiamSeg: Self-Training with Contrastive Learning for Unsupervised Domain Adaptation in Remote Sensing
Bin Wang, Fei Deng, Shuang Wang, Wen Luo, Zhixuan Zhang