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

Active learning theories and methods have been extensively studied inclassical statistical learning settings. However, deep active learning, i.e.,active learning with deep learning models, is usually based on empiricalcriteria without solid theoretical justification, thus suffering from heavydoubts when some of those fail to provide benefits in real applications. Inthis paper, by exploring the connection between the generalization performanceand the training dynamics, we propose a theory-driven deep active learningmethod (dynamicAL) which selects samples to maximize training dynamics. Inparticular, we prove that the convergence speed of training and thegeneralization performance are positively correlated under the ultra-widecondition and show that maximizing the training dynamics leads to bettergeneralization performance. Furthermore, to scale up to large deep neuralnetworks and data sets, we introduce two relaxations for the subset selectionproblem and reduce the time complexity from polynomial to constant. Empiricalresults show that dynamicAL not only outperforms the other baselinesconsistently but also scales well on large deep learning models. We hope ourwork would inspire more attempts on bridging the theoretical findings of deepnetworks and practical impacts of deep active learning in real applications.

Deep Active Learning by Leveraging Training Dynamics