Weichen Zhang, Dong Xu, Wanli Ouyang, Wen Li
The paper introduces the Collaborative and Adversarial Network (CAN) for unsupervised domain adaptation, achieving state-of-the-art results by combining domain-collaborative and adversarial learning strategies.
This research presents a new method called the Collaborative and Adversarial Network (CAN) to improve how machines learn from data when labels aren't available for a new domain. It combines two strategies: one that focuses on learning features specific to the new domain and another that focuses on learning features common to both the old and new domains. By using a self-paced learning approach, the method gradually selects and uses easy-to-identify samples from the new domain to improve learning. The method was tested on several datasets for tasks like object and video action recognition and showed excellent performance compared to previous methods.
