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

A fundamental challenge of over-parameterized deep learning models islearning meaningful data representations that yield good performance on adownstream task without over-fitting spurious input features. This workproposes MaskTune, a masking strategy that prevents over-reliance on spurious(or a limited number of) features. MaskTune forces the trained model to explorenew features during a single epoch finetuning by masking previously discoveredfeatures. MaskTune, unlike earlier approaches for mitigating shortcut learning,does not require any supervision, such as annotating spurious features orlabels for subgroup samples in a dataset. Our empirical results on biasedMNIST, CelebA, Waterbirds, and ImagenNet-9L datasets show that MaskTune iseffective on tasks that often suffer from the existence of spuriouscorrelations. Finally, we show that MaskTune outperforms or achieves similarperformance to the competing methods when applied to the selectiveclassification (classification with rejection option) task. Code for MaskTuneis available at https://github.com/aliasgharkhani/Masktune.

MaskTune: Mitigating Spurious Correlations by Forcing to Explore