Abstract: The performance of machine learning may be expressed as the generalization error.To further study characteristic of generalization error，the decomposition method for generalization error is usually used.Aiming at weighting fusion method and quadratic error loss function，the decomposition equation for generalization error is given.On the basis of this，the decomposition equation for the optimal fusion method is further obtained.

Key words: fusion, decomposition of generalization error, variance, bias, covariance

摘要： 机器学习的性能可以通过泛化误差表达，泛化误差越小，则该学习性能越好，反之则性能越差。为了进一步研究泛化误差的特性，通常采用泛化误差分解的方法。针对加权融合方法，并应用平方误差损失函数，给出了泛化误差的一种分解，在此基础上，进一步获得了加权融合方法的最优泛化误差分解。

关键词: 融合, 泛化误差分解, 方差, 偏差, 协方差