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 networks (DNNs) have recently achieved a great success incomputer vision and several related fields. Despite such progress, currentneural architectures still suffer from catastrophic interference (a.k.a.forgetting) which obstructs DNNs to learn continually. While severalstate-of-the-art methods have been proposed to mitigate forgetting, theseexisting solutions are either highly rigid (as regularization) or time/memorydemanding (as replay). An intermediate class of methods, based on dynamicnetworks, has been proposed in the literature and provides a reasonable balancebetween task memorization and computational footprint. In this paper, we devisea dynamic network architecture for continual learning based on a novelforgetting-free neural block (FFNB). Training FFNB features on new tasks isachieved using a novel procedure that constrains the underlying parameters inthe null-space of the previous tasks, while training classifier parametersequates to Fisher discriminant analysis. The latter provides an effectiveincremental process which is also optimal from a Bayesian perspective. Thetrained features and classifiers are further enhanced using an incremental"end-to-end" fine-tuning. Extensive experiments, conducted on differentchallenging classification problems, show the high effectiveness of theproposed method.

FFNB: Forgetting-Free Neural Blocks for Deep Continual Visual Learning