On-Device Lightweight Transfer Learning for Automotive Intrusion Detection on STM32
Dang-Duy-Tien Vo, Xuan-Bach Nguyen, Quang-Kiet Tran, Hoang-Anh Pham · 2024
This paper presents an emerging automotive approach to cybersecurity by designing a lightweight neural network tailored for vehicle intrusion detection systems (IDS). With the increasing complexity and connectivity of modern automobiles, efficient and effective cybersecurity measures are paramount. Traditional IDS solutions often require considerable computational resources, rendering them impractical for resource-constrained environments like typical microcontrollers in automotive systems. The proposed neural network architecture is not only compact enough to operate within the stringent memory constraints of an STM32 microcontroller but also capable of being trained on-device with minimal computation. This breakthrough allows for unprecedented responsiveness and adaptability in IDS, enabling it to generalize and detect new types of attacks in dynamic automotive contexts. Our approach also includes a lightweight on-device training methodology, ensuring the system remains up-to-date with evolving threat landscapes while maintaining minimal computational requirements.