由于大数据和深度学习发展，人脸识别技术得到飞跃发展，在某些领域已达到商用要求。然而人脸分析中还存在其他问题，如多元人脸属性协同学习。目前人脸属性分析在某些场景下已经取得了较好的性能，却仍旧存在一系列待解决的难点问题。比如多元属性间类别差异大（精细化的年龄估计-101类，性别-2类，种族-3类）以及各个任务识别难度不一，造成多任务训练时不同属性分支收敛性不一致，容易造成某些任务过拟合或者训练不充分，同时也影响各个任务之间的信息交换与迁移。本课题针对上述难点问题，展开基于多元人脸属性协同学习方法的研究。在多任务学习框架指导下，通过构建大规模多元人脸属性数据库，研究多元人脸属性自适应协同学习、非均衡多任务协同学习、以及异源数据联合协同学习，构建稳健的多元人脸属性学习模型，提升人脸属性分析性能。

Due to the development of big data and deep learning, face recognition technology has been well developed and has reached commercial requirements in some areas. However, there are still some challenging problems in face analysis, such as multiple face attribute analysis Co-learning based on still of images. In the analysis of face attributes in non-restricted environment, age, gender and race are the most common research subjects. At present, the analysis of face attributes has achieved good performance in some characteristic scenarios, but there are still a series of difficult problems needed to be solved. For example, there are large differences in the categories of attributes (Fine-grained age estimation-101, Gender-2, Ethnicity-3), resulting in inconsistent convergence of tasks of different tasks in multitasking training, which may easily result in over-fitting or under-fitting. These problem may lead to effect the information change and transfer learning In order to solve the above difficulties, this research focuses on the research of collaborative learning method based on differentiated diversity face attributes. Under the guidance of collaborative learning framework, a robust multi-task face attribute learning model is constructed by studying the multi-task sample imbalance learning, multi-attribute weight adaptive learning and multi-tasking network architecture design to enhance the recognition performance of face attributes.