Deep and Wide Neural Networks for Distinguisher Attacks

The amalgamation of deep and wide neural networks with cryptographic analysis has paved the way for novel approaches in distinguishing and exploiting vulnerabilities within cryptographic algorithms. In this paper, we explore the utilization of deep and wide neural networks to empower distinguisher cryptanalysis attacks, aiming to uncover weaknesses and enhance the security of cryptographic systems. We delve into the perspectives offered by these advanced neural network architectures, elucidating their potential to revolutionize cryptanalysis methodologies through enhanced pattern recognition, automatic feature learning, and adaptive learning capabilities. Additionally, we discuss the challenges inherent in leveraging deep and wide neural networks for distinguisher cryptanalysis, including the need for large-scale datasets, computational resources, and interpretability of neural network decisions. We have generated many datasets using different encryption techniques with differing levels of complexity (key sizes and rounds). We evaluated different artificial intelligence models (MLP, DNN, and CNN) on these datasets to evaluate how effectively they could identify the method of encryption applied. Our studies showed a high success rate, especially for the rectangle encryption technique with 95.21% accuracy.