Int J Performability Eng ›› 2022, Vol. 18 ›› Issue (6): 453-462.doi: 10.23940/ijpe.22.06.p8.453462

Previous Articles    

Encrypted Neural Network

Soumit Mandal, Anindya Mitra*, Sumagna Dey, Pradyut Nath, and Subhrapratim Nath   

  1. Meghnad Saha Institute of Technology, Kolkata, 700150, India
  • Submitted on ; Revised on ; Accepted on
  • Contact: * E-mail address:

Abstract: Neural Networks (NN) are highly valuable tools which are gaining importance in recent years due to their versatility in handling huge data from various sources and gaining meaningful insights on them. They are not only being used in recommender systems but also in real-time applications like auto path detection in autonomous vehicles. However, light is rarely shed on how secure these systems are. With the growing digital world, there are always vulnerabilities and chances to exploit. In this paper, Neural Networks are looked at through the lens of security, and a hybrid algorithm is proposed to enhance the security of neural networks by addressing the weight matrix vulnerability in a neural network. This paper proposes a novel method to involve encryption in the neural network named Encrypted Neural Network (ENN) to prevent any malicious modification of the weight matrix that may harm the training and output of the neural network. The proposed algorithm is tested against how accurately the original neural model is preserved. Finally, the accuracy of the algorithm is checked graphically against the size of the neural model and increasing complexity.

Key words: artificial intelligence, neural networks, security, cryptography, encryption