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

In real-world applications, data do not reflect the ones commonly used forneural networks training, since they are usually few, unlabeled and can beavailable as a stream. Hence many existing deep learning solutions suffer froma limited range of applications, in particular in the case of online streamingdata that evolve over time. To narrow this gap, in this work we introduce anovel and complex setting involving unsupervised meta-continual learning withunbalanced tasks. These tasks are built through a clustering procedure appliedto a fitted embedding space. We exploit a meta-learning scheme thatsimultaneously alleviates catastrophic forgetting and favors the generalizationto new tasks. Moreover, to encourage feature reuse during themeta-optimization, we exploit a single inner loop taking advantage of anaggregated representation achieved through the use of a self-attentionmechanism. Experimental results on few-shot learning benchmarks showcompetitive performance even compared to the supervised case. Additionally, weempirically observe that in an unsupervised scenario, the small tasks and thevariability in the clusters pooling play a crucial role in the generalizationcapability of the network. Further, on complex datasets, the exploitation ofmore clusters than the true number of classes leads to higher results, evencompared to the ones obtained with full supervision, suggesting that apredefined partitioning into classes can miss relevant structural information.

Few-Shot Unsupervised Continual Learning through Meta-Examples