This paper introduces a method for learning multiple visual domains using residual adapter modules. The goal is to develop a neural network architecture that can adapt to various visual domains while maintaining or improving accuracy. The method uses residual adapter modules to dynamically adjust the network for different domains, enabling parameter sharing and efficient learning. The key idea is to reconfigure a deep neural network on the fly to work on different domains as needed. The residual adapter modules allow for a small fraction of model parameters, enabling high-degree parameter sharing between domains. The method is evaluated on the Visual Decathlon Challenge, a benchmark that assesses the ability of representations to capture ten different visual domains and perform well uniformly. The results show that the proposed method outperforms existing approaches in terms of performance and efficiency. The method is also shown to be effective in avoiding forgetting, a common issue in sequential learning. The paper also introduces the Visual Decathlon Challenge, a new benchmark for multiple-domain learning in image recognition. The challenge involves performing well on ten different visual classification tasks, from ImageNet and SVHN to action classification and describable texture recognition. The evaluation metric rewards models that perform well on all domains simultaneously. The results demonstrate that the proposed method achieves high accuracy and efficiency in learning multiple visual domains.This paper introduces a method for learning multiple visual domains using residual adapter modules. The goal is to develop a neural network architecture that can adapt to various visual domains while maintaining or improving accuracy. The method uses residual adapter modules to dynamically adjust the network for different domains, enabling parameter sharing and efficient learning. The key idea is to reconfigure a deep neural network on the fly to work on different domains as needed. The residual adapter modules allow for a small fraction of model parameters, enabling high-degree parameter sharing between domains. The method is evaluated on the Visual Decathlon Challenge, a benchmark that assesses the ability of representations to capture ten different visual domains and perform well uniformly. The results show that the proposed method outperforms existing approaches in terms of performance and efficiency. The method is also shown to be effective in avoiding forgetting, a common issue in sequential learning. The paper also introduces the Visual Decathlon Challenge, a new benchmark for multiple-domain learning in image recognition. The challenge involves performing well on ten different visual classification tasks, from ImageNet and SVHN to action classification and describable texture recognition. The evaluation metric rewards models that perform well on all domains simultaneously. The results demonstrate that the proposed method achieves high accuracy and efficiency in learning multiple visual domains.