局部线性广义单索引回归
Locally Linear Generalized Single Index Regression
Tristan Luca Saidi · Arun Kuchibhotla
中文摘要
针对广义单索引回归模型提出并分析一种估计流程。该框架将响应变量 Y 的条件均值建模为协变量 X 的函数,映射形式为 f∘g,其中 g 为 β-Hölder 连续且 β≥2,f 可为有界变差的不连续函数。所提通用流程对协变量 X 的分布仅要求有界、足够光滑且关于 Lebesgue 测度绝对连续。通过求解局部加权线性最小二乘问题,联合估计梯度向量 v̂(x)≈∇g(x) 与条件均值 μ(x)=f(g(x))。推导了梯度估计 v̂_x 与回归估计 μ̂(x) 的中心极限定理,并给出收敛速率,表明当 β=2 且 f 具有局部 Lipschitz 导数时达到极小极大均方误差速率。与已有方法不同,速度估计结果允许连接函数属于局部有界变差类,从而可处理非光滑甚至不连续的连接函数,同时在几乎每个推断点保持相合性与渐近正态性。最后将模型实例化到 g=Π 为到紧致正则嵌入 C³ 曲线 γ 的最近点投影的情形,并辅以真实与合成数据实验验证方法效用。
关键要点
- 01问题:现有单索引回归方法难以处理不连续或非光滑连接函数,且对协变量分布假设较强。
- 02方法:通过局部加权线性最小二乘联合估计梯度向量与条件均值,仅需协变量有界、光滑且绝对连续。
- 03结果:推导中心极限定理与收敛速率,β=2 时达到极小极大均方误差速率,并支持有界变差连接函数。
- 04应用:实例化到 C³ 曲线最近点投影情形,并用真实与合成数据实验验证。
解读
尚无解读。
原始英文摘要
arXiv:2611.22011v1 Announce Type: cross Abstract: In this work we propose and analyze a procedure for estimating a generalized single-index regression model. This generalized single-index regression framework models the conditional mean of the responses $Y$ as a function of the covariates $X$ where the mapping is assumed to have the form $f \circ g$, where $g$ is $\beta$-H\"older with $\beta \geq 2$ and $f$ can be a discontinuous function of bounded variation. We propose a general procedure that operates under minimal assumptions about the distribution of covariates $X$, requiring only boundedness, sufficient smoothness and absolute continuity with respect to Lebesgue. Our procedure jointly estimates the gradient vector $\hat v(x) \approx \nabla g(x)$ and the conditional mean $\mu(x) = f(g(x))$ by solving a localized weighted linear least squares problem. We derive central limit theorems for gradient vector estimates $\hat{v}_x$ and regression estimates $\hat{\mu}(x)$, and we derive rates of convergence for our approach, illustrating that it indeed achieves a stochastic convergence rate matching the minimax mean squared error rate when $\beta = 2$ and $f$ has a locally Lipschitz derivative. Unlike prior approaches our velocity estimation results accommodate link functions in the local bounded-variation class, allowing for nonsmooth and even discontinuous links while retaining consistency and asymptotic normality at almost every inference point. Finally, we instantiate this model in the situation where $g = \Pi$ is a closest point projection onto a compact and regular embedded $C^3$ curve $\gamma$, a setting recently studied in the literature. We also supplement our theoretical results with experiments on real and synthetic data to demonstrate the utility of our procedure.