The paper presents a novel CNN architecture designed to adapt deep convolutional neural networks (CNNs) to new domains with limited or no labeled data per target category. The approach simultaneously optimizes for domain invariance to facilitate domain transfer and uses a soft label distribution matching loss to transfer information between tasks. The method is evaluated on two standard benchmark visual domain adaptation tasks, the Office dataset and a cross-dataset collection, demonstrating superior performance compared to previous state-of-the-art methods. The authors show that their approach effectively aligns visual domains and transfers semantic structure from the source domain to the target domain, improving classification accuracy in both supervised and semi-supervised settings. The experimental results highlight the effectiveness of domain confusion and soft label loss in enhancing the network's ability to adapt to new domains and transfer task information.The paper presents a novel CNN architecture designed to adapt deep convolutional neural networks (CNNs) to new domains with limited or no labeled data per target category. The approach simultaneously optimizes for domain invariance to facilitate domain transfer and uses a soft label distribution matching loss to transfer information between tasks. The method is evaluated on two standard benchmark visual domain adaptation tasks, the Office dataset and a cross-dataset collection, demonstrating superior performance compared to previous state-of-the-art methods. The authors show that their approach effectively aligns visual domains and transfers semantic structure from the source domain to the target domain, improving classification accuracy in both supervised and semi-supervised settings. The experimental results highlight the effectiveness of domain confusion and soft label loss in enhancing the network's ability to adapt to new domains and transfer task information.