English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Distribution shift (e.g., task or domain shift) in continual learning (CL)usually results in catastrophic forgetting of neural networks. Although it canbe alleviated by repeatedly replaying buffered data, the every-step replay istime-consuming. In this paper, we study which modules in neural networks aremore prone to forgetting by investigating their training dynamics during CL.Our proposed metrics show that only a few modules are more task-specific andsensitively alter between tasks, while others can be shared across tasks ascommon knowledge. Hence, we attribute forgetting mainly to the former and findthat finetuning them only on a small buffer at the end of any CL method canbring non-trivial improvement. Due to the small number of finetuned parameters,such ``Forgetting Prioritized Finetuning (FPF)'' is efficient in computation.We further propose a more efficient and simpler method that entirely removesthe every-step replay and replaces them by only $k$-times of FPF periodicallytriggered during CL. Surprisingly, this ``$k$-FPF'' performs comparably to FPFand outperforms the SOTA CL methods but significantly reduces theircomputational overhead and cost. In experiments on several benchmarks of class-and domain-incremental CL, FPF consistently improves existing CL methods by alarge margin, and $k$-FPF further excels in efficiency without degrading theaccuracy. We also empirically studied the impact of buffer size, epochs pertask, and finetuning modules on the cost and accuracy of our methods.

Does Continual Learning Equally Forget All Parameters?