Label Leakage in Federated Inertial-based Human Activity Recognition

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Bock, Marius, Hopp, Maximilian, Van Laerhoven, Kristof, Moeller, Michael
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866909726114054144
author Bock, Marius
Hopp, Maximilian
Van Laerhoven, Kristof
Moeller, Michael
author_facet Bock, Marius
Hopp, Maximilian
Van Laerhoven, Kristof
Moeller, Michael
contents While prior work has shown that Federated Learning updates can leak sensitive information, label reconstruction attacks, which aim to recover input labels from shared gradients, have not yet been examined in the context of Human Activity Recognition (HAR). Given the sensitive nature of activity labels, this study evaluates the effectiveness of state-of-the-art gradient-based label leakage attacks on HAR benchmark datasets. Our findings show that the number of activity classes, sampling strategy, and class imbalance are critical factors influencing the extent of label leakage, with reconstruction accuracies reaching well-above 90% on two benchmark datasets, even for trained models. Moreover, we find that Local Differential Privacy techniques such as gradient noise and clipping offer only limited protection, as certain attacks still reliably infer both majority and minority class labels. We conclude by offering practical recommendations for the privacy-aware deployment of federated HAR systems and identify open challenges for future research. Code to reproduce our experiments is publicly available via github.com/mariusbock/leakage_har.
format Preprint
id arxiv_https___arxiv_org_abs_2505_20924
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Label Leakage in Federated Inertial-based Human Activity Recognition
Bock, Marius
Hopp, Maximilian
Van Laerhoven, Kristof
Moeller, Michael
Machine Learning
Human-Computer Interaction
While prior work has shown that Federated Learning updates can leak sensitive information, label reconstruction attacks, which aim to recover input labels from shared gradients, have not yet been examined in the context of Human Activity Recognition (HAR). Given the sensitive nature of activity labels, this study evaluates the effectiveness of state-of-the-art gradient-based label leakage attacks on HAR benchmark datasets. Our findings show that the number of activity classes, sampling strategy, and class imbalance are critical factors influencing the extent of label leakage, with reconstruction accuracies reaching well-above 90% on two benchmark datasets, even for trained models. Moreover, we find that Local Differential Privacy techniques such as gradient noise and clipping offer only limited protection, as certain attacks still reliably infer both majority and minority class labels. We conclude by offering practical recommendations for the privacy-aware deployment of federated HAR systems and identify open challenges for future research. Code to reproduce our experiments is publicly available via github.com/mariusbock/leakage_har.
title Label Leakage in Federated Inertial-based Human Activity Recognition
topic Machine Learning
Human-Computer Interaction
url https://arxiv.org/abs/2505.20924