When Can a Byzantine-Affected Federated Intrusion Detector Take a Certified Step? A Client-Indexed Reference Feasibility Study

A federated intrusion detector must balance two competing objectives: rejecting unsupported model updates and continuing to learn attack-discriminative features. We examine when both objectives can be addressed within one conservative optimization rule. We investigate a conditional, client-indexed reference audit for Byzantine-affected federated logistic learning. The server does not know which registered client reports are malicious, but is assumed to receive fresh independent reference samples from every registered client’s pre-corruption distribution. Candidate mean and coordinate-median updates are accepted only when an unknown-honest-subset alignment bound certifies a positive decrease in honest-population logistic risk. In a bounded, label-dependent synthetic intrusion task with ten clients, one −8× Byzantine reporter and eight rounds, five independent seeds each yielded five certified updates with indexed reference data, versus none with a gateway-only certificate. The indexed method achieved mean held-out ROC–AUC 0.7790, attack recall 0.1461, and observed false-positive rate 0.0320 under independently calibrated thresholds. These detection figures are empirical, not certified. Complementary negative controls demonstrate that risk descent does not imply class discrimination and that authenticated reference messages can still contain incorrect labels. Together, the experiments distinguish certificate-backed optimization from empirical detection and identify when independent client-indexed evidence makes a conservative federated detector trainable.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-21
DOI
https://doi.org/10.5281/zenodo.24422490
Primary Topic
Privacy-Preserving Technologies in Data
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

When Can a Byzantine-Affected Federated Intrusion Detector Take a Certified Step? A Client-Indexed Reference Feasibility Study

Md. Amir Khusru Akhtar
Zenodo (CERN European Organization for Nuclear Research)
Privacy-Preserving Technologies in Data
preprint

When Can a Byzantine-Affected Federated Intrusion Detector Take a Certified Step? A Client-Indexed Reference Feasibility Study

Md. Amir Khusru Akhtar
preprint en

Abstract

A federated intrusion detector must balance two competing objectives: rejecting unsupported model updates and continuing to learn attack-discriminative features. We examine when both objectives can be addressed within one conservative optimization rule. We investigate a conditional, client-indexed reference audit for Byzantine-affected federated logistic learning. The server does not know which registered client reports are malicious, but is assumed to receive fresh independent reference samples from every registered client’s pre-corruption distribution. Candidate mean and coordinate-median updates are accepted only when an unknown-honest-subset alignment bound certifies a positive decrease in honest-population logistic risk. In a bounded, label-dependent synthetic intrusion task with ten clients, one −8× Byzantine reporter and eight rounds, five independent seeds each yielded five certified updates with indexed reference data, versus none with a gateway-only certificate. The indexed method achieved mean held-out ROC–AUC 0.7790, attack recall 0.1461, and observed false-positive rate 0.0320 under independently calibrated thresholds. These detection figures are empirical, not certified. Complementary negative controls demonstrate that risk descent does not imply class discrimination and that authenticated reference messages can still contain incorrect labels. Together, the experiments distinguish certificate-backed optimization from empirical detection and identify when independent client-indexed evidence makes a conservative federated detector trainable.

Zenodo (CERN European Organization for Nuclear Research)
Privacy-Preserving Technologies in Data
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon