Abstract:Graph neural networks (GNNs) model representations from networked data and allow for decentralized inference through localized communications. Existing GNN architectures often assume ideal communications and ignore potential channel effects, such as fading and noise, leading to performance degradation in real-world implementation. Considering a GNN implemented over nodes connected through wireless links, this paper conducts a stability analysis to study the impact of channel impairments on the performance of GNNs, and proposes graph neural networks over the air (AirGNNs), a novel GNN architecture that incorporates the communication model. AirGNNs modify graph convolutional operations that shift graph signals over random communication graphs to take into account channel fading and noise when aggregating features from neighbors, thus, improving architecture robustness to channel impairments during testing. We develop a channel-inversion signal transmission strategy for AirGNNs when channel state information (CSI) is available, and propose a stochastic gradient descent based method to train AirGNNs when CSI is unknown. The convergence analysis shows that the training procedure approaches a stationary solution of an associated stochastic optimization problem and the variance analysis characterizes the statistical behavior of the trained model. Experiments on decentralized source localization and multi-robot flocking corroborate theoretical findings and show superior performance of AirGNNs over wireless communication channels.

Privacy-Preserving Decentralized Inference with Graph Neural Networks in Wireless Networks

Devil in Disguise: Breaching Graph Neural Networks Privacy through Infiltration

Privacy-Enhanced Graph Neural Network for Decentralized Local Graphs

Design of Retransmission Mechanism for Decentralized Inference with Graph Neural Networks

A Graph Neural Network Based Decentralized Learning Scheme

A Privacy-Preserving Graph Neural Network for Network Intrusion Detection

Decentralized Graph Neural Network for Privacy-Preserving Recommendation

Graph Neural Networks over the Air for Decentralized Tasks in Wireless Networks

Local Differential Privacy in Graph Neural Networks: a Reconstruction Approach

Industrial Scale Privacy Preserving Deep Neural Network

Releasing Graph Neural Networks with Differential Privacy Guarantees

Towards Private Learning on Decentralized Graphs With Local Differential Privacy

Efficient Privacy Preserving Graph Neural Network for Node Classification

Achieving Adaptive Privacy-Preserving Graph Neural Networks Training in Cloud Environment

Graph neural networks: a survey on the links between privacy and security

Unraveling Privacy Risks of Individual Fairness in Graph Neural Networks

Unveiling the Role of Message Passing in Dual-Privacy Preservation on GNNs

A Privacy-Preserving Subgraph-Level Federated Graph Neural Network via Differential Privacy

Implementing Graph Neural Networks over Wireless Networks Via Over-the-Air Computing: A Joint Communication and Computation Framework

P$$^2$$CG: a Privacy Preserving Collaborative Graph Neural Network Training Framework

Decentralized Statistical Inference with Unrolled Graph Neural Networks