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/

Zendy Plus

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

Zendy Tools

Zendy Open

Generative Adversarial Networks (GANs) are powerful models able to synthesizedata samples closely resembling the distribution of real data, yet thediversity of those generated samples is limited due to the so-called modecollapse phenomenon observed in GANs. Especially prone to mode collapse areconditional GANs, which tend to ignore the input noise vector and focus on theconditional information. Recent methods proposed to mitigate this limitationincrease the diversity of generated samples, yet they reduce the performance ofthe models when similarity of samples is required. To address this shortcoming,we propose a novel method to selectively increase the diversity ofGAN-generated samples. By adding a simple, yet effective regularization to thetraining loss function we encourage the generator to discover new data modesfor inputs related to diverse outputs while generating consistent samples forthe remaining ones. More precisely, we maximise the ratio of distances betweengenerated images and input latent vectors scaling the effect according to thediversity of samples for a given conditional input. We show the superiority ofour method in a synthetic benchmark as well as a real-life scenario ofsimulating data from the Zero Degree Calorimeter of ALICE experiment in LHC,CERN.

Selectively increasing the diversity of GAN-generated samples