Scholarlink

Cyber-physical systems (CPSs) have been used in different domains to enable automation, increase efficiency and effectiveness, and reduce the operational costs of traditional systems. CPSs come with several limitations and requirements that must be considered when designing their application to different domains. Artificial intelligence (AI) can facilitate the optimization of cyber-physical systems’ operation. However, integration of AI functionality into CPS is not easy due to limitations in hardware, software, and flexibility. The main contribution of the present paper is a novel approach for the use of remote AI services in CPSs. By employing zero-knowledge proofs, we protect the privacy of models and data and we can verify the integrity of the operations on the side of the AI service. Our experiments have shown that such an approach is feasible and brings significant offerings, such as verifiable remote AI inference for CPS. Our experiments have shown that currently available zero-knowledge implementations require large proof generation times, which hinder the effective application of remote AI services to real-world CPS.

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

Privacy Preserving and Verifiable Outsourcing of AI Processing for Cyber-Physical Systems

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Georgios Spathoulas,
  • Angeliki Katsika,
  • Georgios Kavallieratos

摘要

Cyber-physical systems (CPSs) have been used in different domains to enable automation, increase efficiency and effectiveness, and reduce the operational costs of traditional systems. CPSs come with several limitations and requirements that must be considered when designing their application to different domains. Artificial intelligence (AI) can facilitate the optimization of cyber-physical systems’ operation. However, integration of AI functionality into CPS is not easy due to limitations in hardware, software, and flexibility. The main contribution of the present paper is a novel approach for the use of remote AI services in CPSs. By employing zero-knowledge proofs, we protect the privacy of models and data and we can verify the integrity of the operations on the side of the AI service. Our experiments have shown that such an approach is feasible and brings significant offerings, such as verifiable remote AI inference for CPS. Our experiments have shown that currently available zero-knowledge implementations require large proof generation times, which hinder the effective application of remote AI services to real-world CPS.