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 Plus annual

Presents the access of premium content as premium feature

Premium Content

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 Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Rehearsal, seeking to remind the model by storing old knowledge in lifelonglearning, is one of the most effective ways to mitigate catastrophicforgetting, i.e., biased forgetting of previous knowledge when moving to newtasks. However, the old tasks of the most previous rehearsal-based methodssuffer from the unpredictable domain shift when training the new task. This isbecause these methods always ignore two significant factors. First, the DataImbalance between the new task and old tasks that makes the domain of old tasksprone to shift. Second, the Task Isolation among all tasks will make the domainshift toward unpredictable directions; To address the unpredictable domainshift, in this paper, we propose Multi-Domain Multi-Task (MDMT) rehearsal totrain the old tasks and new task parallelly and equally to break the isolationamong tasks. Specifically, a two-level angular margin loss is proposed toencourage the intra-class/task compactness and inter-class/task discrepancy,which keeps the model from domain chaos. In addition, to further address domainshift of the old tasks, we propose an optional episodic distillation loss onthe memory to anchor the knowledge for each old task. Experiments on benchmarkdatasets validate the proposed approach can effectively mitigate theunpredictable domain shift.

Multi-Domain Multi-Task Rehearsal for Lifelong Learning