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

The field of meta-learning has seen a dramatic rise in interest in recentyears. In existing meta-learning approaches, learning tasks for trainingmeta-models are usually collected from public datasets, which brings thedifficulty of obtaining a sufficient number of meta-learning tasks with a largeamount of training data. In this paper, we propose a meta-learning approachbased on randomly generated meta-learning tasks to obtain a parametric loss forclassification learning based on big data. The loss is represented by a deepneural network, called meta-loss network (MLN). To train the MLN, we constructa large number of classification learning tasks through randomly generatingtraining data, validation data, and corresponding ground-truth linearclassifier. Our approach has two advantages. First, sufficient meta-learningtasks with large number of training data can be obtained easily. Second, theground-truth classifier is given, so that the difference between the learnedclassifier and the ground-truth model can be measured to reflect theperformance of MLN more precisely than validation accuracy. Based on thisdifference, we apply the evolutionary strategy algorithm to find out theoptimal MLN. The resultant MLN not only leads to satisfactory learning effectson generated linear classifier learning tasks for testing, but also behavesvery well on generated nonlinear classifier learning tasks and various publicclassification tasks. Our MLN stably surpass cross-entropy (CE) and mean squareerror (MSE) in testing accuracy and generalization ability. These resultsillustrate the possibility of achieving satisfactory meta-learning effectsusing generated learning tasks.

Generating meta-learning tasks to evolve parametric loss for classification learning