一种生存数据非线性充分降维子空间估计的新方法

An approach to estimating nonlinear sufficient dimension reduction subspace for censored survival data

给出了一种估计生存数据非线性充分降维子空间的新方法.利用再生核Hilbert空间性质以及双切片思想,建立广义特征谱分解问题与获得充分降维子空间的联系,以此估计生存时间和删失时间的联合非线性降维中心子空间.进一步结合SDR中心子空间的性质,通过联合SDR中心子空间来估计权重函数,在算法实现过程中,利用迭代思想,达到提高估计效率的目的.最后通过数值模拟来说明该方法的优良性.

An approach was proposed to estimating the nonlinear sufficient dimension reduction(SDR)subspace for survival data with censorship.Based on the theory of reproducing kernel Hilbert spaces(RKHS)and the double slicing procedure,the joint nonlinear sufficient dimension reduction central subspace was estimated by means of the generalized eigen-decomposition equation.And the weight function was estimated by the definition and property of SDR central subspace.The efficiency was improved by the iteration method while the algorithm was being implemented.Finally,the performance of the proposed method was illustrated on simulated data.