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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Local explanation methods such as LIME have become popular in MIR as toolsfor generating post-hoc, model-agnostic explanations of a model'sclassification decisions. The basic idea is to identify a small set ofhuman-understandable features of the classified example that are mostinfluential on the classifier's prediction. These are then presented as anexplanation. Evaluation of such explanations in publications often resorts toaccepting what matches the expectation of a human without actually being ableto verify if what the explanation shows is what really caused the model'sprediction. This paper reports on targeted investigations where we try to getmore insight into the actual veracity of LIME's explanations in an audioclassification task. We deliberately design adversarial examples for theclassifier, in a way that gives us knowledge about which parts of the input arepotentially responsible for the model's (wrong) prediction. Asking LIME toexplain the predictions for these adversaries permits us to study whether localexplanations do indeed detect these regions of interest. We also look atwhether LIME is more successful in finding perturbations that are moreprominent and easily noticeable for a human. Our results suggest that LIME doesnot necessarily manage to identify the most relevant input features and henceit remains unclear whether explanations are useful or even misleading.

On the Veracity of Local, Model-agnostic Explanations in Audio Classification: Targeted Investigations with Adversarial Examples