Machine-Learning Models in Cyber-Security (2025 Survey): Transformers, Graph Nets & Autoencoders for Malware, Phishing & IPS
Machine Learning Models for Cybersecurity Tasks Cryptography Models Differential Neural Cryptanalysis (ResNet CNN) Architecture & Task: Uses deep neural networks (often residual CNNs) as distinguisher models to aid cryptanalysis. For example, Gohr’s pioneering 2019 work trained a residual CNN to distinguish encrypted ciphertext pairs from random, improving classical differential cryptanalysis on block ciphers. Recent models incorporate advanced layers like residual connections and gated linear units (GLUs) to predict key bits from known plaintext–ciphertext pairs. The neural net takes pairs (or structures) of data and learns to infer partial key information or identify non-random patterns. ...