Scholarlink

<p>The use of public cloud platforms to store the data from a Wireless Sensor Network (WSN) introduces significant privacy and security challenges. While conventional encryption schemes ensure confidentiality during data transmission and storage, they do not support direct computation over encrypted data in untrusted environments. Homomorphic Encryption (HE) addresses this limitation by enabling algebraic operations on ciphertexts, thereby allowing secure data processing on the cloud without decryption and leveraging the server’s computational resources. This work presents a Privacy-Preserving Machine Learning (PPML) framework that performs sign detection and comparison in the encrypted domain. The proposed PPML model is an optimized Neural Network Model for Classification (NNMC) that operates entirely on encrypted data within the public cloud. This NNMC is well-suited for classification tasks, as the responses for sign detection and comparison are discrete values, <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\( - \)</EquationSource> </InlineEquation>1, 0, 1, irrespective of the numerical magnitudes of the operands. Randomized encryption is employed to protect against Chosen Plaintext Attacks (CPA). Key challenges associated with NNMC, such as model accuracy, scalability, and limitations imposed by the HE, are addressed systematically. Experimental results demonstrate that with appropriate hyperparameter tuning, the PPML driven NNMC achieves a classification accuracy of 99.9999%.</p>

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

Sign detection and comparison for wireless sensor data based on subtractive homomorphism and privacy-preserving deep learning model

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月10日

查看全文
  • Neeta B. Malvi,
  • N. Shylashree

摘要

The use of public cloud platforms to store the data from a Wireless Sensor Network (WSN) introduces significant privacy and security challenges. While conventional encryption schemes ensure confidentiality during data transmission and storage, they do not support direct computation over encrypted data in untrusted environments. Homomorphic Encryption (HE) addresses this limitation by enabling algebraic operations on ciphertexts, thereby allowing secure data processing on the cloud without decryption and leveraging the server’s computational resources. This work presents a Privacy-Preserving Machine Learning (PPML) framework that performs sign detection and comparison in the encrypted domain. The proposed PPML model is an optimized Neural Network Model for Classification (NNMC) that operates entirely on encrypted data within the public cloud. This NNMC is well-suited for classification tasks, as the responses for sign detection and comparison are discrete values, \( - \) 1, 0, 1, irrespective of the numerical magnitudes of the operands. Randomized encryption is employed to protect against Chosen Plaintext Attacks (CPA). Key challenges associated with NNMC, such as model accuracy, scalability, and limitations imposed by the HE, are addressed systematically. Experimental results demonstrate that with appropriate hyperparameter tuning, the PPML driven NNMC achieves a classification accuracy of 99.9999%.