资源受限的多视图学习技术研究

With the developments of multi-source data collection techniques, it becomes particularly important that designing multi-view learning approaches for real applications. We focus on four issues on multi-view learning approaches under resource limited environments in this project: 1. Multi-view learning models are relatively complex and require more storages, how to perform multi-view learning based on shared subspace approaches for environments where storages are limited? 2. Feature extraction is time consuming in multi-view learning, how to extract features in each view sequentially for adapting to the limitations of response time in multi-view learning? 3. Multi-view information can be incomplete for many reasons at times, e.g., failures of sensors, how to learn with these incomplete multi-view data and perform missing view imputation? 4. Some multi-view applications forbid cross-view data access, how to improve the classification performance for each view with information privacy and independency preserved in environments where data sharing is restricted? As an application of the solutions to these four issues, we will also build a prototype system. It is expected to publish 8-10 high quality papers on prominent international journals, conferences and top native journals, apply 2-3 patents, and supervise 4-6 graduate students in this project.

随着多源数据收集能力的提高，将多视图学习技术用于实际场景变得尤为重要。本项目对多视图学习在资源受限现实场景下所面临的四个问题进行研究：首先，多视图模型相对复杂并将耗费较大的存储空间，如何进行基于共享子空间的多视图学习以适应存储空间受限的环境？第二，多视图对象描述复杂，特征抽取时间长，如何进行视图按序抽取以提出适于响应时间受限情况下的多视图学习方法？第三，相对于单视图数据，多视图数据容易出现视图缺失现象，例如传感设备故障导致的视图缺失，如何利用不全面的视图信息进行学习和填补缺失视图？第四，在部分多视图应用场景下，视图之间信息无法互访，如何在保持信息隐私和独立的前提下提升多视图分类器的性能，以适应视图分享受限的环境？本项目将为上述问题提供解决方案并研制原型系统，发表国际期刊/会议和国内一级学报论文8-10篇，申请专利2-3项，培养4-6名研究生。

随着多源数据收集能力的提高，将多视图学习技术用于实际场景变得尤为重要。本项目对资源受限现实场景下的多视图学习进行研究，分别提出了适于存储空间受限的多视图学习方法、适于响应时间受限的多视图学习方法、适于采样条件受限的多视图学习方法和适于信息分享受限的多视图学习方法，并研制原型系统。基于研究成果发表论文20篇，其中期刊论文4篇，包括机器学习顶级期刊MLJ、IEEE TPAMI 2篇、IEEE TKDE 1篇；会议论文16篇，包括机器学习顶级会议NeurIPS和ICML 4篇、其他CCF-A类会议AAAI/IJCAI/KDD 8篇，1篇论文获国际会议论文奖；申请国家发明专利4项，培养毕业研究生8名，2人获优博。