Scholarlink

In the realm of Connected and Autonomous Vehicles (CAVs), securing Deep Neural Networks (DNNs) poses a critical challenge, particularly in ensuring confidentiality and integrity amidst potential attacks and unauthorized access. This paper presents ELITE, a distributed approach enhancing security and computational efficiency for V2X applications. ELITE partitions DNNs used in Roadside Units (RSUs) between the RSU (edge device) equipped with a Hardware Security Module (HSM) and the cloud, with specific layers processed within a Trusted Execution Environment (TEE). By leveraging public-key encryption protocols, ELITE facilitates secure model distribution and inference. Real-time integrity verification is conducted with the Zymkey HSM at the RSU while selective processing on the AWS Nitro Enclave offers enhanced security. Experimental results show that ELITE reduces inference time by 45.73% compared to solely cloud-based processing while maintaining high accuracy with the correct encryption key. This distributed approach balances security and efficiency, mitigating risks of unauthorized access and preserving model integrity within DNN-enabled V2X roadside services.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

ELITE: Efficient and Secure Machine Learning for Intelligent Perception in Smart Road Infrastructure

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Swathi Kumar Vembu,
  • Trupil Limbasiya,
  • Anupam Chattopadhyay

摘要

In the realm of Connected and Autonomous Vehicles (CAVs), securing Deep Neural Networks (DNNs) poses a critical challenge, particularly in ensuring confidentiality and integrity amidst potential attacks and unauthorized access. This paper presents ELITE, a distributed approach enhancing security and computational efficiency for V2X applications. ELITE partitions DNNs used in Roadside Units (RSUs) between the RSU (edge device) equipped with a Hardware Security Module (HSM) and the cloud, with specific layers processed within a Trusted Execution Environment (TEE). By leveraging public-key encryption protocols, ELITE facilitates secure model distribution and inference. Real-time integrity verification is conducted with the Zymkey HSM at the RSU while selective processing on the AWS Nitro Enclave offers enhanced security. Experimental results show that ELITE reduces inference time by 45.73% compared to solely cloud-based processing while maintaining high accuracy with the correct encryption key. This distributed approach balances security and efficiency, mitigating risks of unauthorized access and preserving model integrity within DNN-enabled V2X roadside services.