Effects of Un targeted Adversarial Attacks on Deep Learning Methods

DEĞİRMENCİ, Elif; Özçelik, İlker; Yazıcı, Ahmet

doi:10.1109/iscturkey56345.2022.9931786

Cited by 2 publications

(2 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For example, (Hoffmann et al 2021) investigates ProtoP-Nets interpretability and discovers a semantic gap between similarity in input and latent space. At the same time, (Etmann et al 2019;Zhang and Zhu 2019;Tsipras et al 2018) highlight connections between the explainability of machine learning models and their adversarial robustness. However, according to our knowledge, no systematic benchmark has been proposed for a comprehensive comparison of prototypical parts-based models, such as the one we propose.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Interpretability Benchmark for Evaluating Spatial Misalignment of Prototypical Parts Explanations

Sacha,

Jura,

Rymarczyk

et al. 2024

AAAI

View full text Add to dashboard Cite

Prototypical parts-based networks are becoming increasingly popular due to their faithful self-explanations. However, their similarity maps are calculated in the penultimate network layer. Therefore, the receptive field of the prototype activation region often depends on parts of the image outside this region, which can lead to misleading interpretations. We name this undesired behavior a spatial explanation misalignment and introduce an interpretability benchmark with a set of dedicated metrics for quantifying this phenomenon. In addition, we propose a method for misalignment compensation and apply it to existing state-of-the-art models. We show the expressiveness of our benchmark and the effectiveness of the proposed compensation methodology through extensive empirical studies.

show abstract

Section: Related Workmentioning

confidence: 99%

“…Untargeted attacks are another type of adversarial attack, aiming to manipulate pixel intensities to reduce the confidence of the original class prediction until it is no longer the dominant one (Degirmenci, Ozcelik, and Yazici 2022).…”

Section: Related Workmentioning

confidence: 99%