no code implementations • 5 Dec 2021 • Bilan Lai, Haoran Xiang, Furao Shen
We perform extensive experiments to prove that pruning based on the influence function using the idea of ensemble learning will be much more effective than just focusing on error reconstruction.