Noise-Aware Misclassification Attack Detection in Collaborative DNN Inference

2026-03-18

AI Summary

This work addresses the challenge of detecting stealthy misclassifications caused by malicious data injection during collaborative deep neural network inference between resource-constrained devices and edge servers, particularly under environmental noise that obscures adversarial perturbations. To enhance detection robustness, the paper proposes a noise-aware semi-gray-box anomaly detection framework that leverages a variational autoencoder (VAE) to model feature distributions and explicitly disentangles environmental noise from adversarial perturbations. Evaluated across multiple mainstream image classification models, the approach achieves up to 90% AUROC while significantly reducing false positive rates, demonstrating its effectiveness in realistic noisy settings. The study also highlights the inherent challenges posed by high noise levels and feature similarity, which can impede reliable anomaly detection.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Data Mining & Knowledge Management: Anomaly/Outlier Detection

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

Collaborative inference of object classification Deep neural Networks (DNNs) where resource-constrained end-devices offload partially processed data to remote edge servers to complete end-to-end processing, is becoming a key enabler of edge-AI. However, such edge-offloading is vulnerable to malicious data injections leading to stealthy misclassifications that are tricky to detect, especially in the presence of environmental noise. In this paper, we propose a semi-gray-box and noise- aware anomaly detection framework fueled by a variational autoencoder (VAE) to capture deviations caused by adversarial manipulation. The proposed framework incorporates a robust noise-aware feature that captures the characteristic behavior of environmental noise to improve detection accuracy while reducing false alarm rates. Our evaluation with popular object classification DNNs demonstrate the robustness of the proposed detection (up to 90% AUROC across DNN configurations) under realistic noisy conditions while revealing limitations caused by feature similarity and elevated noise levels.

Problem

• misclassification attack
• collaborative DNN inference
• adversarial detection
• environmental noise
• edge-AI

Innovation

• noise-aware detection
• collaborative DNN inference
• variational autoencoder
• misclassification attack
• edge-AI security