Pruning Deep Neural Networks via the Marchenko-Pastur Distribution

Arxiv CS.LG (machine learning)

Paper proposes pruning DNNs using Marchenko-Pastur distribution from random matrix theory, requiring very short post-pruning fine-tuning. Main contribution is accuracy retention with minimal calibration and fine-tuning schedules.

deep-learning

Arxiv CS.LG (machine learning)·Leonid BerlyandTheo BourdaisHouman OwhadYitzchak Shmalo

8 days ago