Scholarlink

With the development of MLaaS in recent years, the integrity and data security have also faced challenges. Ensuring the consumers receive correct services and safeguarding the data security is important. In this paper, we designed algorithms for each module, using depthwise separable convolution to optimize the convolution module and using the minimum polynomial method to optimize the circuit design of the activation function. Additionally, this paper proposes the SSHC algorithm for deeper network connections and combines blockchain to verify the correctness of the proofs. The experimental results demonstrate that our approach reduces proof time by 54.6% and storage by 58.1% compared to the original solution, while maintaining high efficiency and practicality.

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

Validating the Integrity for Deep Learning Models Based on Zero-Knowledge Proof and Blockchain

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Qianyi Zhan,
  • Yuanyuan Liu,
  • Zhenping Xie,
  • Yuan Liu

摘要

With the development of MLaaS in recent years, the integrity and data security have also faced challenges. Ensuring the consumers receive correct services and safeguarding the data security is important. In this paper, we designed algorithms for each module, using depthwise separable convolution to optimize the convolution module and using the minimum polynomial method to optimize the circuit design of the activation function. Additionally, this paper proposes the SSHC algorithm for deeper network connections and combines blockchain to verify the correctness of the proofs. The experimental results demonstrate that our approach reduces proof time by 54.6% and storage by 58.1% compared to the original solution, while maintaining high efficiency and practicality.