Multilinear Subspace Regression and Its Application in BCI.

本项目主要研究两个方面：1)多路数据（张量）分析方法。我们提出并发展了许多张量数据的监督学习方法，这些方法可以对多维结构化数据进行多元线性回归或分类。此外，为了捕捉张量数据的非线性关系，我们提出了一组可以处理张量值输入的核函数，这为将核机应用于张量空间打开了大门。2)脑机接口与脑信号分析。我们开发了一种情感脑机接口（BCI），使用情绪面孔作为刺激。另一方面，我们将提出的张量技术广泛应用于脑信号分析，在脑信号解码、特征提取和erp分类方面表现出显著的性能提升。

This project focuses on two aspects: 1) Multiway data (tensor) analysis methods. We proposed and developed many supervised learning methods for tensor data, which can perform multilinear regression or classification on multi-dimensional structured data. In addition, in order to capture the nonlinear relations of tensor data, we proposed a family of kernel functions that can handle tensor-valued inputs, which opens a door for applying kernel machines to tensor space. 2) Brain computer interface and brain signal analysis. We developed an affective brain computer interface (BCI) using emotional faces as stimuli. On the other hand, we extensively applied our proposed tensor techniques for analyzing brain signals, which have shown significant improvement of performance in terms of decoding of brain signals, feature extractions and classifications of ERPs.