김승호, 김형식,

Vol. 34, No. 5, pp. 865-873, 10월. 2024
10.13089/JKIISC.2024.34.5.865, Full Text:
Keywords: Privacy-Preserving Machine Learning, homomorphic encryption, Cryptographic API Misuse Detection, Convolutional Neural Network
Abstract

In this study, we propose a privacy-preserving cryptographic API misuse detection framework utilizing homomorphic encryption. The proposed framework is designed to effectively detect cryptographic API misuse while maintaining data confidentiality. We employ a Convolutional Neural Network (CNN)-based detection model and optimize its structure to ensure high accuracy even in an encrypted environment. Specifically, to enable efficient homomorphic operations, we leverage depth-wise convolutional layers and a cubic activation function to secure non-linearity, enabling effective misuse detection on encrypted data. Experimental results show that the proposed model achieved a high F1-score of 0.978, and the total execution time for the homomorphically encrypted model was 11.20 seconds, demonstrating near real-time processing efficiency. These findings confirm that the model offers excellent security and accuracy even when operating in a homomorphic encryption environment.

Statistics
Cite this article