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

Training datasets are crucial for convolutional neural network-basedalgorithms, which directly impact their overall performance. As such, using awell-structured dataset that has minimum level of bias is always desirable. Inthis paper, we proposed a novel, domain-independent approach, calledscore-based resampling (SBR), to locate the under-represented samples of theoriginal training dataset based on the model prediction scores obtained withthat training set. In our method, once trained, we use the same CNN model toinfer on its own training samples, obtain prediction scores, and based on thedistance between predicted and ground-truth, we identify samples that are faraway from their ground-truth and augment them in the original training set. Thetemperature term of the Sigmoid function is decreased to better differentiatescores. For experimental evaluation, we selected one Kaggle dataset for genderclassification. We first used a CNN-based classifier with relatively standardstructure, trained on the training images, and evaluated on the providedvalidation samples of the original dataset. Then, we assessed it on a totallynew test dataset consisting of light male, light female, dark male, and darkfemale groups. The obtained accuracies varied, revealing the existence ofcategorical bias against certain groups in the original dataset. Subsequently,we trained the model after resampling based on our proposed approach. Wecompared our method with a previously proposed variational autoencoder (VAE)based algorithm. The obtained results confirmed the validity of our proposedmethod regrading identifying under-represented samples among original datasetto decrease categorical bias of classifying certain groups. Although tested forgender classification, the proposed algorithm can be used for investigatingdataset structure of any CNN-based tasks.

Dataset Bias Mitigation Through Analysis of CNN Training Scores