Application of Deep Learning in the Optimization of AES Encryption Algorithm

Abstract

With the rapid development of information technology, data security issues have become increasingly prominent. As one of the most widely used symmetric encryption algorithms currently, AES (Advanced Encryption Standard) plays a crucial role in ensuring data confidentiality. However, with the improvement of computing power and the explosive growth of data volume, The traditional AES algorithm faces growing challenges in terms of performance and security. In scenarios where the resources of Internet of Things (IoT) devices are limited, the high computational complexity of AES encryption introduces latency in device response, making it difficult to meet real-time requirements. Meanwhile, the breakthrough of quantum computing technology threatens AES with fast decryption, and new side-channel attacks (e.g., power analysis attacks, electromagnetic radiation analysis attacks) jeopardize its security. This paper aims to explore the application of deep learning technology in the optimization of the AES encryption algorithm. By deeply analyzing the bottleneck problems of the AES algorithm in practical applications, targeted optimization schemes are proposed. Specifically, this paper utilizes the powerful learning ability and feature extraction ability of deep learning models to optimize the key generation, encryption round function and other key components of the AES algorithm, in order to solve the problems of low efficiency when the traditional AES algorithm processes large-scale data and its limited resistance to new attacks. Through the construction of optimization models based on Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN) and Generative Adversarial Networks (GAN), and a large number of simulation experiments and practical tests, the research results show that the deep learning-based optimization method can effectively improve the encryption speed of the AES algorithm, with the encryption efficiency increased by 35% on the ARM Cortex-M4 chip. At the same time, it significantly enhances its security, and the ability to resist differential attacks and linear attacks is increased by more than 40%, providing new ideas and methods for the further development of the AES algorithm in practical applications. Specifically, by introducing LSTM to model the AES key expansion process, the Shannon entropy of the round key is increased from 122.3 bits in the traditional algorithm to 145.6 bits, effectively enhancing the randomness of the key. In terms of resisting differential attacks, the differential characteristic probability of the optimized AES algorithm is reduced from 0.12 to 0.07, significantly improving the security of the algorithm.