基于高斯过程动态系统的多输出时序数据分类与回归

Gaussian process dynamical systems are among the main methods for modeling sequential data. With the advent of big data, large-scale extension of the models becomes an important research frontier in machine learning. We will investigate the classification and regression problems for large-scale sequential data, based on the lately proposed dependent multi-output Gaussian process dynamical systems (DMGPDS), and further extend the dynamical systems to dynamical regression models to take into account the dynamics in some regression problems. The specific research topics include classification models based on DMGPDS, large-scale regression models based on DMGPDS, large-scale classification models based on DMGPDS, and dependent multi-output Gaussian process dynamical regression models and their large-scale extensions. Through theoretical research and experimental evaluations on these topics, we can enhance the understanding about DMGPDS, and put forward reasonable and effective models and algorithms for large-scale sequential data classification and regression. The outcome of this project can break through the limitations of complex Gaussian process dynamical systems for dealing with large-scale data, and inspire and promote the research on Gaussian process dynamical systems.

高斯过程动态系统是时序数据建模的主流方法之一。随着大数据时代的到来，模型的大规模扩展构成了机器学习领域中的前沿研究方向。我们将基于最近提出的依赖多输出高斯过程动态系统，开展对大规模时序数据的分类与回归问题的研究，并将动态系统模型扩展为动态回归模型以兼顾某些回归问题中潜在的动态性。具体研究内容包括：依赖多输出高斯过程动态系统分类模型；基于依赖多输出高斯过程动态系统的大规模数据回归；基于依赖多输出高斯过程动态系统的大规模数据分类；依赖多输出高斯过程动态回归模型及其大规模扩展。通过对这些内容进行理论研究与实验验证，可以增强对依赖多输出高斯过程动态系统特点的认识，提出合理有效的模型和算法以完成大规模时序数据的分类与回归任务。本项目的研究成果将突破复杂高斯过程动态系统处理大规模时序数据的局限，对高斯过程动态系统的研究与应用起到启发与促进作用。

高斯过程动态系统是时序数据建模的主流方法之一。随着大数据时代的到来，模型的大规模扩展构成了机器学习领域中的前沿研究方向。我们基于最近提出的依赖多输出高斯过程动态系统，开展了对时序数据的分类与回归问题的研究，并将动态系统模型扩展为动态回归模型以兼顾某些回归问题中潜在的动态性。具体研究内容包括：混合高斯过程与深度高斯过程模型；依赖多输出高斯过程动态系统分类模型；基于依赖多输出高斯过程动态系统的大规模数据回归与分类；依赖多输出高斯过程动态回归模型及其扩展；多视图支持向量机学习方法。通过对这些内容进行理论研究与实验验证，增强了对高斯过程动态系统等模型特点的认识，提出了合理有效的模型和算法。本项目的研究成果对高斯过程的研究与应用起到了促进作用。