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

Interpretation and improvement of deep neural networks relies on betterunderstanding of their underlying mechanisms. In particular, gradients ofclasses or concepts with respect to the input features (e.g., pixels in images)are often used as importance scores or estimators, which are visualized insaliency maps. Thus, a family of saliency methods provide an intuitive way toidentify input features with substantial influences on classifications orlatent concepts. Several modifications to conventional saliency maps, such asRectified Gradients and Layer-wise Relevance Propagation (LRP), have beenintroduced to allegedly denoise and improve interpretability. While visuallycoherent in certain cases, Rectified Gradients and other modified saliency mapsintroduce a strong input bias (e.g., brightness in the RGB space) because ofinappropriate uses of the input features. We demonstrate that dark areas of aninput image are not highlighted by a saliency map using Rectified Gradients,even if it is relevant for the class or concept. Even in the scaled images, theinput bias exists around an artificial point in color spectrum. Ourmodification, which simply eliminates multiplication with input features,removes this bias. This showcases how a visual criteria may not align with trueexplainability of deep learning models.

Input Bias in Rectified Gradients and Modified Saliency Maps