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

Deep Neural Network (DNN) has achieved great success on datasets of closedclass set. However, new classes, like new categories of social media topics,are continuously added to the real world, making it necessary to incrementallylearn. This is hard for DNN because it tends to focus on fitting to new classeswhile ignoring old classes, a phenomenon known as catastrophic forgetting.State-of-the-art methods rely on knowledge distillation and data replaytechniques but still have limitations. In this work, we analyze the causes ofcatastrophic forgetting in class incremental learning, which owes to threefactors: representation drift, representation confusion, and classifierdistortion. Based on this view, we propose a two-stage learning framework witha fixed encoder and an incrementally updated prototype classifier. The encoderis trained with self-supervised learning to generate a feature space with highintrinsic dimensionality, thus improving its transferability and generality.The classifier incrementally learns new prototypes while retaining theprototypes of previously learned data, which is crucial in preserving thedecision boundary.Our method does not rely on preserved samples of old classes,is thus a non-exemplar based CIL method. Experiments on public datasets showthat our method can significantly outperform state-of-the-art exemplar-basedmethods when they reserved 5 examplers per class, under the incremental settingof 10 phases, by 18.24% on CIFAR-100 and 9.37% on ImageNet100.

Class Incremental Learning with Self-Supervised Pre-Training and Prototype Learning