Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
J. He and L. Bai are with Institute of Intelligent Information Processing, Shanxi University, Taiyuan, 030006, China (Corresponding author: Liang Bai) Email: hejinli@sxu.edu.cn, bailiang@sxu.edu.cn X. Yang is with Alliance Manchester Business School, The University of Manchester, Manchester, M13 9PL, UK. Email: xian.yang@manchester.ac.uk
2
0
3
Rehearsal, a cornerstone of continual learning, can paradoxically *hurt* a model's ability to adapt to new tasks, challenging the conventional wisdom that more rehearsal is always better.
Forget about complex continual learning strategies: this work shows that simply tuning the *magnitude* of parameter updates can significantly reduce forgetting and improve generalization.