Partial least squares for serially dependent data

该项目侧重于偏最小二乘（PLS）-一种为病态线性回归模型开发的正则化最小二乘回归。我们寻求这种技术的扩展，使处理串行相关的数据。这种扩展必须追求的PLS算法的线性和非线性版本。因此，主要的重点应该是渐近理论（渐近分布和一致性的估计，均方预测误差等）。所开发的方法将被应用于预测的一个特定的生物功能的蛋白质的基础上集体原子运动的这种蛋白质，无论是观察随着时间的推移和已知的高度自相关。

This projects focuses on the partial least squares (PLS) – a type of regularized least squares regression developed for ill-conditioned linear regression models. We seek for the extension of this technique that enables to handle serially dependent data. This extension has to be pursued for both – linear and nonlinear – versions of the PLS algorithm. Thereby, the main focus should be on the asymptotic theory (asymptotic distribution and consistency of the estimators, mean squared prediction error, etc.). The developed method will be applied to the prediction of a specific biological function of a protein based on the collective atomic motion of this protein, both observed over time and known to be highly autocorrelated.