HEKAN:A new neural network for encrypted inference
Yuwei Xu, Han Liu, Zhongyu Hu, Xiaodong Li, Yimeng Dou, Shaofei Xu · 2025
With the rapid development of neural network, the contradiction between improving the performance of network models and ensuring data privacy has become increasingly prominent. This paper proposes a novel neural network model, Homomorphic Encryption Kolmogorov-Arnold Networks(HEKAN), which deeply integrates Fully Homomorphic Encryption technology with Kolmogorov-Arnold Networks(KAN). Our salient advantage resides in the successful implementation of a neural network model, herein referred to as the HEKAN model, which takes advantage of the KAN model’s few parameters, high accuracy, and strong characterization capabilities to make the cryptographic inference process more interpretable, and faster, also enhances security and efficiency in inference process, thereby laying a robust foundation for secure applications within the realm of artificial intelligence.