Jul 2026
Rethinking Transfer in Continual Learning: A Replay-Based Realisation
TSR consistently improves forward transfer while maintaining stability, outperforming existing replay baselines across heterogeneous and homogeneous task streams and argues for treating transfer as a first-class objective of continual learning, to be understood before it is engineered.
Yang Meng, Zhenyang Liu, Zhuokai Zhao et al.
· arXiv.org · 0 citations