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 Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Social media platforms prevent malicious activities by detecting harmfulcontent of posts and comments. To that end, they employ large-scale deep neuralnetwork language models for sentiment analysis and content understanding. Somemodels, like BERT, are complex, and have numerous parameters, which makes themexpensive to operate and maintain. To overcome these deficiencies, industryexperts employ a knowledge distillation compression technique, where adistilled model is trained to reproduce the classification behavior of theoriginal model. The distillation processes terminates when the distillationloss function reaches the stopping criteria. This function is mainly designedto ensure that the original and the distilled models exhibit alikeclassification behaviors. However, besides classification accuracy, there areadditional properties of the original model that the distilled model shouldpreserve to be considered as an appropriate abstraction. In this work, weexplore whether distilled TinyBERT models preserve confidence values of theoriginal BERT models, and investigate how this confidence preservation propertycould guide tuning hyperparameters of the distillation process.

Confidence Preservation Property in Knowledge Distillation Abstractions