Scalability in Ill-Posed Machine Learning Problems: Bridging Least Squares Methods With (Non-)Convex Algorithms
MCML Authors
Claudio Mayrink Verdun
Dr.
* Former Member
→ Group Felix Krahmer
* Former PI
Abstract
Claudio Mayrink Verdun
Dr.
* Former Member
→ Group Felix Krahmer
* Former PI
Abstract
We introduce novel algorithms to address some challenges in machine learning, including ill-conditioned low-rank matrix retrieval, constrained least squares, and high-dimensional regression with unknown noise. By bridging least squares with modern (non-)convex optimization, our methods achieve scalability, data efficiency, and robustness. We provide theoretical guarantees with minimal assumptions and numerically validate their performance.
BibTeXKey: Ver22