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

Feature attribution methods have become a staple method to disentangle thecomplex behavior of black box models. Despite their success, some scholars haveargued that such methods suffer from a serious flaw: they do not allow areliable interpretation in terms of human concepts. Simply put, visualizing anarray of feature contributions is not enough for humans to conclude somethingabout a model's internal representations, and confirmation bias can trick usersinto false beliefs about model behavior. We argue that a structured approach isrequired to test whether our hypotheses on the model are confirmed by thefeature attributions. This is what we call the "semantic match" between humanconcepts and (sub-symbolic) explanations. Building on the conceptual frameworkput forward in Cin\`a et al. [2023], we propose a structured approach toevaluate semantic match in practice. We showcase the procedure in a suite ofexperiments spanning tabular and image data, and show how the assessment ofsemantic match can give insight into both desirable (e.g., focusing on anobject relevant for prediction) and undesirable model behaviors (e.g., focusingon a spurious correlation). We couple our experimental results with an analysison the metrics to measure semantic match, and argue that this approachconstitutes the first step towards resolving the issue of confirmation bias inXAI.

Fixing confirmation bias in feature attribution methods via semantic match