Cornell University
Smoothness-Penalized Deconvolution: Rates of Convergence, Choice of Tuning Parameter, and Inference
Abstract
dc:description.abstractThis work addresses the deconvolution problem of estimating a square-integrable probability density from observations contaminated with additive measurement errors having a known density. The estimator begins with a density estimate of the contaminated observations and minimizes a reconstruction error penalized by an integrated squared m-th derivative. Theory for deconvolution has mainly focused on kernel- or wavelet-based techniques, but other methods including spline-based techniques and this smoothness-penalized estimator have been found to outperform kernel methods in simulation studies. This work fills in some of the gaps in theory by proving rates of convergence for the smoothness-penalized estimator and its spline approximant. We contribute to the practical use of the estimator by developing an unbiased estimator of its risk, along with a method for using that estimated risk to choose the tuning parameter; this outperforms the SURE-based tuning parameter method that has been proposed for a similar estimator. Finally, we develop methods for constructing bias-corrected pointwise confidence intervals and assess the coverage properties in a simulation study, finding that they have uniformly lower coverage error than the naive normal-theory intervals.
Degree
thesis:*- Name thesis:degree_name
- Ph. D., Statistics
- Level thesis:degree_level
- Doctor of Philosophy
- Discipline thesis:degree_discipline
- Statistics
- Grantor
- Cornell University
- Year dc:date.issued
- 2023
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Kent, David
- Chair dc:contributor.chair
-
- Ruppert, David
- Committee members dc:contributor.committeemember
-
- Kato, Kengo
- Loredo, Thomas
Subjects
dc:subject × 6Rights
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Dc Identifier Bibid
- 16454749
- Dc Identifier Other
-
ProQuest Submission ID: 14041
ProQuest Publication ID: 30690402