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

How to enable learnability for new classes while keeping the capability wellon old classes has been a crucial challenge for class incremental learning.Beyond the normal case, long-tail class incremental learning and few-shot classincremental learning are also proposed to consider the data imbalance and datascarcity, respectively, which are common in real-world implementations andfurther exacerbate the well-known problem of catastrophic forgetting. Existingmethods are specifically proposed for one of the three tasks. In this paper, weoffer a unified solution to the misalignment dilemma in the three tasks.Concretely, we propose neural collapse terminus that is a fixed structure withthe maximal equiangular inter-class separation for the whole label space. Itserves as a consistent target throughout the incremental training to avoiddividing the feature space incrementally. For CIL and LTCIL, we further proposea prototype evolving scheme to drive the backbone features into our neuralcollapse terminus smoothly. Our method also works for FSCIL with only minoradaptations. Theoretical analysis indicates that our method holds the neuralcollapse optimality in an incremental fashion regardless of data imbalance ordata scarcity. We also design a generalized case where we do not know the totalnumber of classes and whether the data distribution is normal, long-tail, orfew-shot for each coming session, to test the generalizability of our method.Extensive experiments with multiple datasets are conducted to demonstrate theeffectiveness of our unified solution to all the three tasks and thegeneralized case.

Neural Collapse Terminus: A Unified Solution for Class Incremental Learning and Its Variants