Kunpeng Li

Random Sparse Networks Training with Sharpness-Aware Regularization.

This study examines a complementary perspective on LTH and proposes the Dual Lottery Ticket Hypothesis (DLTH), enabling general training for sparse networks and introduces Sharpness-Aware Random Sparse Network Transformation (SA-RST), which uses a regularization term to extrude information from weights slated for maski...