United States Department of Transportation
ROSA P serves as an archival repository of USDOT-published products including scientific findings, journal articles, guidelines, recommendations, or other information authored or co-authored by USDOT or funded partners.
As a repository, ROSA P retains documents in their original published format to ensure public access to scientific information.
i
A Software Tool for Securing Deep Learning Against Adversarial Attacks for CAVs
-
2024-11-01
-
[PDF-759.99 KB]
Details:
-
Creators:
-
Corporate Creators:
-
Corporate Contributors:
-
Subject/TRT Terms:
-
Resource Type:
-
Geographical Coverage:
-
Edition:Final Report (October 2023 – October 2024)
-
Corporate Publisher:
-
Abstract:This project focuses on the technological transfer of a robust perception algorithm previously developed to mitigate adversarial attacks, transforming it into a practical software tool with an intuitive interface. The initiative builds upon the prior project, Securing Deep Learning against Adversarial Attacks for Connected and Automated Vehicles, which successfully introduced a deep ensemble network combining discriminative and generative models to counter adversarial examples. This innovative approach utilized a causal latent graph embedded in a Bayesian model to estimate adversarial perturbations, demonstrating superior accuracy and robustness when trained solely on clean data. The current project advances this work by prioritizing usability and accessibility, emphasizing the development of a graphical user interface (GUI) to facilitate the generation, training, and testing of adversary-resilient neural networks. The anticipated outcome is a tool that democratizes access to robust AI systems, enabling diverse users to enhance the security of perception systems in various applications, regardless of their expertise in deep learning.
-
Format:
-
Collection(s):
-
Main Document Checksum:
-
Download URL:
-
File Type:
