Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Otte, Clemens, Yang, Yinchong, Oswan, Danny Benlin
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2408.00783
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866913455780397056
author Otte, Clemens
Yang, Yinchong
Oswan, Danny Benlin
author_facet Otte, Clemens
Yang, Yinchong
Oswan, Danny Benlin
contents The paper proposes a new testing approach for Deep Neural Networks (DNN) using gradient-free optimization to find perturbation chains that successfully falsify the tested DNN, going beyond existing grid-based or combinatorial testing. Applying it to an image segmentation task of detecting railway tracks in images, we demonstrate that the approach can successfully identify weaknesses of the tested DNN regarding particular combinations of common perturbations (e.g., rain, fog, blur, noise) on specific clusters of test images.
format Preprint
id arxiv_https___arxiv_org_abs_2408_00783
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Data-driven Verification of DNNs for Object Recognition
Otte, Clemens
Yang, Yinchong
Oswan, Danny Benlin
Computer Vision and Pattern Recognition
The paper proposes a new testing approach for Deep Neural Networks (DNN) using gradient-free optimization to find perturbation chains that successfully falsify the tested DNN, going beyond existing grid-based or combinatorial testing. Applying it to an image segmentation task of detecting railway tracks in images, we demonstrate that the approach can successfully identify weaknesses of the tested DNN regarding particular combinations of common perturbations (e.g., rain, fog, blur, noise) on specific clusters of test images.
title Data-driven Verification of DNNs for Object Recognition
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2408.00783