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

Continual Learning (CL) aims to incrementally update a trained model on newtasks without forgetting the acquired knowledge of old ones. Existing CLmethods usually reduce forgetting with task priors, \ie using task identity ora subset of previously seen samples for model training. However, these methodswould be infeasible when such priors are unknown in real-world applications. Toaddress this fundamental but seldom-studied problem, we propose a Prior-FreeContinual Learning (PFCL) method, which learns new tasks without knowing thetask identity or any previous data. First, based on a fixed single-headarchitecture, we eliminate the need for task identity to select thetask-specific output head. Second, we employ a regularization-based strategyfor consistent predictions between the new and old models, avoiding revisitingprevious samples. However, using this strategy alone often performs poorly inclass-incremental scenarios, particularly for a long sequence of tasks. Byanalyzing the effectiveness and limitations of conventionalregularization-based methods, we propose enhancing model consistency with anauxiliary unlabeled dataset additionally. Moreover, since some auxiliary datamay degrade the performance, we further develop a reliable sample selectionstrategy to obtain consistent performance improvement. Extensive experiments onmultiple image classification benchmark datasets show that our PFCL methodsignificantly mitigates forgetting in all three learning scenarios.Furthermore, when compared to the most recent rehearsal-based methods thatreplay a limited number of previous samples, PFCL achieves competitiveaccuracy. Our code is available at: https://github.com/visiontao/pfcl

Prior-Free Continual Learning with Unlabeled Data in the Wild