Original title: Privacy, Robustness, and Fairness Trade-offs in Federated Intrusion Detection: Geometric Indistinguishability at the Aggregation Interface
Adrita Rahman Tory, ABM Shawkat Ali, Md Abu Layek, Khondokar Fida Hasan · September 2026
Computer networks use machine learning to spot cyberattacks. Federated learningA method where multiple devices train a shared model without sending their data to a central server. lets different systems train one shared defense model together without sharing private network data. However, real-world systems need three things at once: strict data privacy, defense against malicious participants trying to sabotage training, and the ability to spot rare cyberattacks. Past research assumed you could easily combine all three goals. The researchers tested this idea by combining privacy noise with a defense filter on an intrusion dataset. To protect privacy, systems add random mathematical noise to their updates. The researchers discovered that this noise scatters the data updates. Because of this dispersion, the defense filter cannot tell the difference between rare attack signals and malicious sabotage. As a result, the system accidentally filters out rare attacks. Combining privacy and defense disproportionately harms the detection of uncommon cyber threats.
Imagine a group of friends collaborating on a group project without meeting in person. To keep their ideas private, they add random distractions to their shared notes. But if one friend has a unique, detailed idea that only appears once, the added noise might erase it entirely, making the group miss a valuable contribution. In this paper, privacy noise in federated learning can similarly wash out the faint signals of rare attacks.
This affects cybersecurity teams using collaborative learning to detect intrusions without sharing data. If privacy protections hide rare attacks, those threats could slip through unnoticed.
The study provides initial evidence based on one dataset and specific attack types. It focuses on label-flip and model-poisoning, so other attacks might behave differently. The findings suggest the need for joint evaluation but do not propose a complete solution.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Privacy, Robustness, and Fairness Trade-offs in Federated Intrusion Detection: Geometric Indistinguishability at the Aggregation Interface — Adrita Rahman Tory, ABM Shawkat Ali, Md Abu Layek et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.