1 code implementation • 14 Dec 2023 • Zhenrong Liu, Yang Li, Yi Gong, Yik-Chung Wu
This approach optimizes network parameters in the null space of the past tasks' feature representation matrix to guarantee the stability.
Continual Learning