Derives distribution-dependent generalization bounds for tuning L1/L2 regularization in multi-task linear regression that do not degrade with feature dimension d under sub-Gaussian i.i.d. assumptions and are sharper than prior uniform bounds for large d.
Dispersion for data-driven algorithm design, online learning, and private optimization
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Distribution-dependent Generalization Bounds for Tuning Linear Regression Across Tasks
Derives distribution-dependent generalization bounds for tuning L1/L2 regularization in multi-task linear regression that do not degrade with feature dimension d under sub-Gaussian i.i.d. assumptions and are sharper than prior uniform bounds for large d.