Neural networks based domain name generation

Domain generation algorithm (DGA) is used by botnets to build a stealthy command and control (C&C) communication channel between the C&C server bots. A DGA can periodically produce large number of pseudo-random algorithmically generated domains (AGDs), few which direct bots server. AGD detection algorithms provide lightweight, promising solution in response existing techniques. In constantly evolving attacker–defender game, attackers may seek more advanced techniques gain better chance evading defenders. this paper, we propose new DGA, namely neural networks-based domain name (NDG) architecture. NDG based on variational autoencoder (VAE), where encoder decoder networks use stacked gated convolutional (GCNNs) learn contextual structure hierarchically. experimentally validated using set state-of-the-art algorithms. The DGAs different classes following taxonomy are benchmark NDG. shows best overall anti-detection performance among all tested DGAs. We also demonstrate that effective benchmarking