CAREER: Harness the Big Data via Large-Scale Lifelong Learning

大数据基础设施和算法基础的最新进展释放了大量数据，这些数据被收集并存储在世界各地的分布式数据中心。这些海量数据集的不断增加导致了许多内在相关的机器学习任务。因此，在过去的十年里，迁移学习范式被发展起来，以执行任务之间的知识转移，以提高任务的泛化性能。该项目将开发一套大规模终身学习方法，以应对大数据知识转移带来的重大挑战。该项目中开发的算法和工具将直接影响生物医学信息学和智能交通系统，因为它们将被用于根据电子医疗记录建立个性化预测模型，以及从大数据中建立交通状态模型。这个项目的成功将被用来开发一种新的课程，将研究纳入课堂，并为来自代表性不足群体的学生提供参与机器学习研究的机会。大数据的速度、容量、可变性和多样性等特性给传统的终身学习方法带来了巨大的挑战。该项目将通过以下方式推动终身学习：(1)开发一个分布式终身学习框架，以实现大规模分布式数据集上的在线知识转移；(2)设计有效的方法来跟踪任务关系中的时间漂移，并通过交互转移利用人类知识；(3)研究使分布式终身学习能够处理来自特征空间和学习任务的异构性的策略。该项目的成果将为许多大数据分析提供一个随时可用的大规模终身学习框架，从而对大数据理论和算法基础产生立竿见影的影响。所有调查结果、出版物、软件和数据将在项目网站上公开提供：http://jiayuzhou.github.io/projects/career.This奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Recent advances in big data infrastructures and algorithm foundations have unleashed a torrent of data being collected and stored in distributed data centers all over the world. The ever-increasing availability of these massive datasets leads to many machine learning tasks that are inherently related. Therefore transfer learning paradigms have been developed in the past decade to perform knowledge transfer among tasks to improve their generalization performance. This project will develop a suite of large-scale lifelong learning methods to address significant challenges from knowledge transfer on big data. The algorithms and tools developed in this project will directly impact biomedical informatics and intelligent transportation systems, as they will be used to build personalized predictive models from electronic medical records and traffic state models from big traffic data. The success of this project will be used to develop a new curriculum that incorporates research into the classroom and provides students from under-represented groups with opportunities to participate in machine learning research. The properties of velocity, volume, variability, and variety that characterize big data have imposed significant challenges in the traditional lifelong learning approaches. This project will advance lifelong learning by (1) developing a distributed life-long learning framework to enable online knowledge transfer on large-scale distributed datasets; (2) designing effective methods to track temporal drifting in the task relationship, and leverage human knowledge via interactive transfer; and (3) investigating strategies that enable the distributed life-long learning to handle heterogeneities from both feature spaces and learning tasks. The results of this project will have an immediate and strong impact on Big Data theoretical and algorithmic foundations, by enabling a large-scale lifelong learning framework readily available for many Big Data analytics. All findings, publications, software, and data will be made publicly available at the project website: http://jiayuzhou.github.io/projects/career.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.