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) has shown promising results and comparableperformance to learning at once in a fully supervised manner. However, CLstrategies typically require a large number of labeled samples, making theirreal-life deployment challenging. In this work, we focus on semi-supervisedcontinual learning (SSCL), where the model progressively learns from partiallylabeled data with unknown categories. We provide a comprehensive analysis ofSSCL and demonstrate that unreliable distributions of unlabeled data lead tounstable training and refinement of the progressing stages. This problemseverely impacts the performance of SSCL. To address the limitations, wepropose a novel approach called Dynamic Sub-Graph Distillation (DSGD) forsemi-supervised continual learning, which leverages both semantic andstructural information to achieve more stable knowledge distillation onunlabeled data and exhibit robustness against distribution bias. Firstly, weformalize a general model of structural distillation and design a dynamic graphconstruction for the continual learning progress. Next, we define a structuredistillation vector and design a dynamic sub-graph distillation algorithm,which enables end-to-end training and adaptability to scale up tasks. Theentire proposed method is adaptable to various CL methods and supervisionsettings. Finally, experiments conducted on three datasets CIFAR10, CIFAR100,and ImageNet-100, with varying supervision ratios, demonstrate theeffectiveness of our proposed approach in mitigating the catastrophicforgetting problem in semi-supervised continual learning scenarios.

Dynamic Sub-graph Distillation for Robust Semi-supervised Continual Learning