Trustworthy and Scalable Nonconvex Statistical Estimation for Sample-Starved Multi-Modal Data Models
Abstract
Major Goals: The overarching goal of this research program is to investigate reliable, model-agnostic, and provably accurate information processing and inference procedures in the challenging data-hungry regime (i.e. when the number of samples at hand is not necessarily much larger than the underlying degrees of freedom of the data models). Particular emphasis is placed on multi-modal and heterogeneous data models, including the cases where (1) the sensing units are heterogeneous so that the samples acquired might have drastically different characteristics, or (2) the acquired samples are driven simultaneously by multiple important sources and we are asked to disentangle these sources. New insights and novel techniques from high-dimensional statistics, mathematical optimization, and statistical learning theory will be developed to meet the research objectives. Throughout the proposal, for concreteness, we frame our discussions in a few stylized problems such as mixed regression, blind deconvolution and de-mixing, spectral learning, tensor completion, etc. We believe that the techniques to be developed in this research program are broadly applicable to other foundational problems of critical values to the defense applications.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 31, 2021
- Accession Number
- AD1190355
Entities
People
- Yuxin Chen
Organizations
- Princeton University