New Research Links Covariate Dependence to Multiple Descent in Regression

A new paper published on arXiv explores over-parameterized linear regression, relaxing common assumptions about independent covariates and non-degenerate covariance matrices. The research demonstrates that degeneracy and dependence can lead to multiple descent phenomena, identifying specific configurations where variance becomes singular. These findings are characterized using a novel graph representation of the variance profile, with maximum matchings and the Dulmage--Mendelsohn decomposition pinpointing the locations of these singular variance peaks. AI

IMPACT Provides theoretical insights into regression models that could inform future AI algorithm development.

RANK_REASON The cluster contains a single academic paper published on arXiv detailing novel mathematical findings in statistics. [lever_c_demoted from research: ic=1 ai=0.4]

Read on arXiv stat.ML →

• alphaXiv
• arXiv
• CatalyzeX Code Finder for Papers
• CORE Recommender
• DagsHub
• Dulmage--Mendelsohn decomposition
• Gotit.pub
• Hugging Face
• Influence Flower
• ScienceCast
• The Zero Pattern of a Design Matrix Drives Multiple Descent in Over-parameterized Regression

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →