Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Detection of Global Anomalies on Distributed IoT Edges with Device-to-Device Communication

Hideya Ochiai, Riku Nishihata, Eisuke Tomiyama, Yuwei Sun, Hiroshi Esaki

TL;DR

This work tackles the challenge of high-frequency anomaly detection in distributed IoT edge networks by proposing WAFL-Autoencoder, a fully distributed autoencoder trained through Wireless Ad Hoc Federated Learning (WAFL) with device-to-device communication. It formalizes Global versus Local Anomalies and introduces a distributed thresholding mechanism that aggregates per-device thresholds to detect global anomalies without sharing raw data. The approach demonstrates that distributed autoencoders can learn global normal features across non-IID edge data and detect global anomalies with high true positive rates and low false positives on standard benchmark datasets. The findings suggest a scalable, privacy-preserving, low-communication-path solution for edge-based anomaly detection in IoT environments, with potential applications to electric monitoring, motion sensing, and imaging at the edge.

Abstract

Anomaly detection is an important function in IoT applications for finding outliers caused by abnormal events. Anomaly detection sometimes comes with high-frequency data sampling which should be carried out at Edge devices rather than Cloud. In this paper, we consider the case that multiple IoT devices are installed in a single remote site and that they collaboratively detect anomalies from the observations with device-to-device communications. For this, we propose a fully distributed collaborative scheme for training distributed anomaly detectors with Wireless Ad Hoc Federated Learning, namely "WAFL-Autoencoder". We introduce the concept of Global Anomaly which sample is not only rare to the local device but rare to all the devices in the target domain. We also propose a distributed threshold-finding algorithm for Global Anomaly detection. With our standard benchmark-based evaluation, we have confirmed that our scheme trained anomaly detectors perfectly across the devices. We have also confirmed that the devices collaboratively found thresholds for Global Anomaly detection with low false positive rates while achieving high true positive rates with few exceptions.

Detection of Global Anomalies on Distributed IoT Edges with Device-to-Device Communication

TL;DR

This work tackles the challenge of high-frequency anomaly detection in distributed IoT edge networks by proposing WAFL-Autoencoder, a fully distributed autoencoder trained through Wireless Ad Hoc Federated Learning (WAFL) with device-to-device communication. It formalizes Global versus Local Anomalies and introduces a distributed thresholding mechanism that aggregates per-device thresholds to detect global anomalies without sharing raw data. The approach demonstrates that distributed autoencoders can learn global normal features across non-IID edge data and detect global anomalies with high true positive rates and low false positives on standard benchmark datasets. The findings suggest a scalable, privacy-preserving, low-communication-path solution for edge-based anomaly detection in IoT environments, with potential applications to electric monitoring, motion sensing, and imaging at the edge.

Abstract

Anomaly detection is an important function in IoT applications for finding outliers caused by abnormal events. Anomaly detection sometimes comes with high-frequency data sampling which should be carried out at Edge devices rather than Cloud. In this paper, we consider the case that multiple IoT devices are installed in a single remote site and that they collaboratively detect anomalies from the observations with device-to-device communications. For this, we propose a fully distributed collaborative scheme for training distributed anomaly detectors with Wireless Ad Hoc Federated Learning, namely "WAFL-Autoencoder". We introduce the concept of Global Anomaly which sample is not only rare to the local device but rare to all the devices in the target domain. We also propose a distributed threshold-finding algorithm for Global Anomaly detection. With our standard benchmark-based evaluation, we have confirmed that our scheme trained anomaly detectors perfectly across the devices. We have also confirmed that the devices collaboratively found thresholds for Global Anomaly detection with low false positive rates while achieving high true positive rates with few exceptions.
Paper Structure (14 sections, 5 equations, 3 figures, 2 tables)

This paper contains 14 sections, 5 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. WAFL-Autoencoder for Anomaly Detection
  4. Definition of Terms
  5. Global Anomaly
  6. Implicit Data Category and Class
  7. Distributions of Train, Validation, and Test Data
  8. Aggregation of Distributed Models with Device-to-Device Communication
  9. Finding Anomaly Threshold with Device-to-Device Communication
  10. Evaluation
  11. Experiment Setting
  12. Reconstructed Images
  13. Detection of Global Anomalies
  14. Conclusion

Figures (3)

  • Figure 1: Local anomaly and global anomaly examples. Local anomaly is something rare to the device, but well-known to others. Global anomaly is rare to every device, which is difficult to detect with low false-positive rates.
  • Figure 2: Reconstructed images of (a) legitimate inputs and (b) global anomaly inputs at epoch = 0, 100, 500, and 5000 (at device 3). The model could reconstruct minor legitimate class images successfully while keeping global anomaly outputs unconstructed.
  • Figure 3: Positive rates of anomaly detection to the test data at device 0 for the models from epoch 0 to 5000. MNIST are false positive rates (FPR). Noisy-, Occluded-, Fashion-, and Kuzhushiji-MNIST are true positive rates (TPR). The number $y$ in MNIST-$y$ indicates FPR for class $y$ test samples.