2 papers
cs.LG2026
Gradient Descent with Large Step Size Restores Symmetry in Deep Linear Networks with Multi-Pathway
Hee-Sung Kim, Sungyoon Lee
Recent analyses of multi-pathway Deep Linear Networks use Gradient Flow to predict a "winner-takes-all" specialization in which path symmetry breaks and each feature concentrates i…
cs.LG2026
Inconsistency-Aware Minimization: Improving Generalization with Unlabeled Data
Hee-Sung Kim, Hyeonseong Kim, Sungyoon Lee
Estimating the generalization gap and developing optimization methods that improve generalization are crucial for deep learning models, for both theoretical understanding and pract…