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 popularity of pretrained language models in natural language processingsystems calls for a careful evaluation of such models in down-stream tasks,which have a higher potential for societal impact. The evaluation of suchsystems usually focuses on accuracy measures. Our findings in this paper callfor attention to be paid to fairness measures as well. Through the analysis ofmore than a dozen pretrained language models of varying sizes on two toxic textclassification tasks (English), we demonstrate that focusing on accuracymeasures alone can lead to models with wide variation in fairnesscharacteristics. Specifically, we observe that fairness can vary even more thanaccuracy with increasing training data size and different randominitializations. At the same time, we find that little of the fairnessvariation is explained by model size, despite claims in the literature. Toimprove model fairness without retraining, we show that two post-processingmethods developed for structured, tabular data can be successfully applied to arange of pretrained language models. Warning: This paper contains samples ofoffensive text.

Your fairness may vary: Pretrained language model fairness in toxic text classification