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

Deep neural networks are susceptible to catastrophic forgetting when trainedon sequential tasks. Various continual learning (CL) methods often rely onexemplar buffers or/and network expansion for balancing model stability andplasticity, which, however, compromises their practical value due to privacyand memory concerns. Instead, this paper considers a strict yet realisticsetting, where the training data from previous tasks is unavailable and themodel size remains relatively constant during sequential training. To achievesuch desiderata, we propose a conceptually simple yet effective method thatattributes forgetting to layer-wise parameter overwriting and the resultingdecision boundary distortion. This is achieved by the synergy between two keycomponents: HSIC-Bottleneck Orthogonalization (HBO) implements non-overwrittenparameter updates mediated by Hilbert-Schmidt independence criterion in anorthogonal space and EquiAngular Embedding (EAE) enhances decision boundaryadaptation between old and new tasks with predefined basis vectors. Extensiveexperiments demonstrate that our method achieves competitive accuracyperformance, even with absolute superiority of zero exemplar buffer and 1.02xthe base model.

Towards Continual Learning Desiderata via HSIC-Bottleneck Orthogonalization and Equiangular Embedding