ML Security PapersAdVAR-DNN: Adversarial Misclassification Attack on Collaborative DNN Inference
Shima Yousefi , Motahare Mounesan , Saptarshi Debroy
Published on arXiv
2508.01107
Input Manipulation Attack
OWASP ML Top 10 — ML01
Key Finding
Achieves high attack success rate with little to no detection probability against popular object classification DNNs on CIFAR-100 in a black-box collaborative inference setting.
AdVAR-DNN
Novel technique introduced
In recent years, Deep Neural Networks (DNNs) have become increasingly integral to IoT-based environments, enabling realtime visual computing. However, the limited computational capacity of these devices has motivated the adoption of collaborative DNN inference, where the IoT device offloads part of the inference-related computation to a remote server. Such offloading often requires dynamic DNN partitioning information to be exchanged among the participants over an unsecured network or via relays/hops, leading to novel privacy vulnerabilities. In this paper, we propose AdVAR-DNN, an adversarial variational autoencoder (VAE)-based misclassification attack, leveraging classifiers to detect model information and a VAE to generate untraceable manipulated samples, specifically designed to compromise the collaborative inference process. AdVAR-DNN attack uses the sensitive information exchange vulnerability of collaborative DNN inference and is black-box in nature in terms of having no prior knowledge about the DNN model and how it is partitioned. Our evaluation using the most popular object classification DNNs on the CIFAR-100 dataset demonstrates the effectiveness of AdVAR-DNN in terms of high attack success rate with little to no probability of detection.
Key Contributions
- AdVAR-DNN: a VAE-based adversarial attack exploiting unsecured intermediate feature transmission in collaborative DNN inference between IoT devices and edge/cloud servers
- Classifier-assisted model inference to gather cut-point and architecture information without prior knowledge of the target DNN
- Untraceable adversarial sample generation that achieves high misclassification with low detection probability on CIFAR-100
🛡️ Threat Analysis
AdVAR-DNN generates adversarial/manipulated intermediate feature representations using a VAE to cause misclassification at inference time. While it targets intermediate layers rather than raw inputs, the core contribution is an evasion attack causing misclassification — the hallmark of ML01. The collaborative inference channel is the delivery mechanism, not a supply chain attack.