USI - Email

Archive / INF Seminars / INF_2025_03_06_MasoudTehrani

USI - Email

	Adversarial Attacks on Deep Learning-Based Perception in Autonomous Vehicles: Designing and Evaluating System-Level Failures

Chair: Carmen Armenti

Thursday

06.03

USI Campus EST, Room D1.13

16:30 - 17:30


	Masoud Jamshidiyan Tehrani Università della Svizzera italiana

Abstract: The rise of deep learning and its remarkable success in perception tasks, such as object recognition and classification, has made it an integral component of complex systems, including autonomous vehicles (AVs). However, deep learning models are highly susceptible to mispredictions when small, adversarial perturbations are introduced into their input. These mispredictions can occur in real-world scenarios and potentially escalate from localized errors to critical system-wide failures. In recent years, an increasing number of research efforts have explored adversarial attacks on autonomous vehicles, specifically targeting their deep learning-based perception systems. These attacks manipulate elements of the vehicle’s environment, with their impact measured in terms of the system-level failures they induce. In this project, we focus on deliberately causing system-level failures in AVs by exploiting vulnerabilities in their deep learning components. Our approach involves reviewing existing literature, designing novel adversarial attacks, and evaluating their effectiveness in a simulated environment. By exposing these weaknesses, this research aims to contribute to the development of more robust deep learning models, ultimately enhancing the safety of autonomous vehicles.

Biography: Masoud holds a bachelor’s degree in computer engineering and a master’s degree in cybersecurity from the University of Shiraz. His strong passion for computer science, combined with exceptional problem-solving skills, earned him the top-ranked student position with the highest grades in both programs. He is currently pursuing a Ph.D. at Università della Svizzera Italiana under the supervision of Paolo Tonella and Jinhan Kim. As part of the Sec4AI4Sec European project, his research focuses on attacking the perception systems of autonomous vehicles. Beyond academia, Masoud enjoys challenging himself through competitive video gaming and learning new languages. He is also an active volunteer, organizing events for exchange students. With his strong social skills, he helps them make the most of their stay in Lugano, fostering connections and creating memorable experiences.

Chair: Carmen Armenti

*************************
In February 2019, the Software Institute started its SI Seminar Series. Every Thursday afternoon, a researcher of the Institute will publicly give a short talk on a software engineering argument of their choice. Examples include, but are not limited to novel interesting papers, seminal papers, personal research overview, discussion of preliminary research ideas, tutorials, and small experiments.
On our YouTube playlist you can watch some of the past seminars. On the SI website you can find more details on the next seminar, the upcoming seminars, and an archive of the past speakers.