文献翻译：
In this paper we propose a new flexible functional regression approach that gener-alizes the popular functional linear model. We adopt an additive decomposition of r which uses projections in the space of the functional predictor. Note that projection has a direct relation to predictability and interpretability of the model. We shall keep this feature by defining the most predictable direction and generalize the linear re-lationship to include nonlinear or nonparametric relationship. The additive structure provides a flexible alternative to the nonparametric approach having often reason-
able interpretations. Since our approach extends the PPR principle to the functional context, it can be named Functional Projection Pursuit Regression (FPPR).
本文提出一个新的复杂函数回归，推广了函数型线性回归。采取可加分解，使用在函数型自变量空间上的投影。注意投影对模型的预测和理解都有关。定义最大预测方向，推广线性关系，使得包含非线性或者非参数关系。可加结构提供了非参数途径一种复杂的选择，常常有很好的解释。因为这个方法推广了ppr准则，他可以被命名为函数型ppr.
2 Most predictive directions and components
The aim is to find the principal (projection pursuit) direction θ∗j along which to project X and the function g∗j that explain the most variability of Y with respect to X. The functions g∗j are called the ridge functions in the multivariate statistics literature. The pairs (θ∗j ,g∗j ) are defined as the solutions of the successive minimization problems as explained below
目的是找到主要的方向，使得把x投影，使得函数能够解释y关于x的变化。函数被称为岭函数。数对定义为连续最小问题的解。
3 Construction of the estimates
3.1 Estimation of the one-dimensional functional components
3.2 Estimates of the most predictive directions
3.3 How to choose the number of components
4 Computational issues
able interpretations. Since our approach extends the PPR principle to the functional context, it can be named Functional Projection Pursuit Regression (FPPR).
本文提出一个新的复杂函数回归，推广了函数型线性回归。采取可加分解，使用在函数型自变量空间上的投影。注意投影对模型的预测和理解都有关。定义最大预测方向，推广线性关系，使得包含非线性或者非参数关系。可加结构提供了非参数途径一种复杂的选择，常常有很好的解释。因为这个方法推广了ppr准则，他可以被命名为函数型ppr.
2 Most predictive directions and components
The aim is to find the principal (projection pursuit) direction θ∗j along which to project X and the function g∗j that explain the most variability of Y with respect to X. The functions g∗j are called the ridge functions in the multivariate statistics literature. The pairs (θ∗j ,g∗j ) are defined as the solutions of the successive minimization problems as explained below
目的是找到主要的方向，使得把x投影，使得函数能够解释y关于x的变化。函数被称为岭函数。数对定义为连续最小问题的解。
3 Construction of the estimates
3.1 Estimation of the one-dimensional functional components
3.2 Estimates of the most predictive directions
3.3 How to choose the number of components
4 Computational issues
