LASIGE researchers have published a new study examining the threat of adversarial machine learning to network intrusion detection systems in the top 10% journal Computer Science Review. The paper, titled “Understanding the Adversary: A Survey of Adversarial Machine Learning in Network Intrusion Detection”, results from the collaboration of PhD student  Allan S. Espindola and integrated researchers António Casimiro and Pedro M. Ferreira, with Pontifical Catholic University of Paraná researchers Altair O.Santin and Eduardo K. Viegas.

The survey analyses 94 quality-screened studies published between 2022 and 2025 and organises them through a threat model and hierarchical taxonomy. This framework makes it easier to compare assumptions about adversaries, attack and defense methods, and evaluation practices across the field. The analysis reveals that many experiments still assume adversaries with extensive knowledge or evaluate attacks without establishing whether they could be carried out on real network traffic. The study offers a roadmap for making future research more realistic, reproducible, and operationally viable.

The paper is available here. The data and supporting research materials are also publicly available here.