A unified spectral analysis of high-dimensional linear regression: debiasing and thresholding
高维线性回归的统一谱分析:去偏与阈值化AI 解读 ↓
本页整理公开书目信息,不生成学术质量评分。AI 解读根据题名及所注明的摘要或原文片段生成,未完成全文核验,可能有误;具体假设、证明与结论请核对原文。摘要保留来源文本,未经模型改写。
DOI 首次登记不等于接收或发表日期。部分更正、评论等条目没有独立摘要;缺少可靠材料时保留待补状态。
本页为预印本版本;arXiv 收录、作者填写的期刊信息以及作者既有发表经历,都不单独证明本篇已接收。作者发表记录按 OpenAlex 身份与完整姓名、本站覆盖期刊 ISSN 及正式卷期核对,只展示已核实例,档案仍可能误合并。
引用量采用 OpenAlex 单条关联记录,不合计多个记录;关联记录可能包含正式发表版本,不等于此版本单独引用。缺失不视为零,引用指标不代表论文质量。
关注作者按书目中的完整姓名匹配,同名和姓名变体需人工辨别。相关论文先按共享细分方向和具体方法排列,再参考通用方法与发表日期。
AI 根据所提供的论文材料归类:摘要聚焦高维线性回归的谱分析、去偏与阈值化理论,属于高维统计与计量方法,未涉及所提供领域细分中的因果推断、面板、时间序列、资产定价等具体方向,故不选择。
高维线性回归的统一谱分析:去偏与阈值化AI 解读 ↓
正在核对关联版本…
正在核对对应版本的资源…
We use spectral analysis to establish a unified theory of debiasing, thresholding, and simultaneous inference for high-dimensional linear regression with ill-conditioned designs. The methods include Ridge regression, spectral cut-off, and iterative and dynamical regression procedures. A common debiasing procedure uses two applications of the original linear algorithm with fixed parameters. Spectral analysis shows that debiasing squares the residual, while implementations based on matrix-vector products require no explicit spectral decomposition. Under explicit design, signal, and moment conditions, we establish maximum-norm consistency, recovery of the population threshold set, and error bounds for coefficient and noise-variance estimation. We derive Gaussian approximations for maximum coordinate errors and, under full column rank and deterministic sequences of regression parameters and thresholds, prove bootstrap validity for simultaneous confidence regions. Under additional regularity and gain conditions, we relate noise-free bias rates to best-worst-case error rates. Sparse and non-sparse regression experiments examine the effects of debiasing and thresholding under specified tuning rules. An elliptic inverse source problem extends the numerical assessment to reconstruction and empirical simultaneous coverage.