Task-Oriented Communication for Graph Data: A Graph Information Bottleneck Approach

TL;DR

This paper introduces a method to extract a smaller, task-focused subgraph that maintains key information while reducing communication overhead, and proposes the VQ-GIB mechanism, integrating vector quantization to convert subgraph representations into a discrete codebook sequence, compatible with existing digital communication systems.

Abstract

Graph data, essential in fields like knowledge representation and social networks, often involves large networks with many nodes and edges. Transmitting these graphs can be highly inefficient due to their size and redundancy for specific tasks. This paper introduces a method to extract a smaller, task-focused subgraph that maintains key information while reducing communication overhead. Our approach utilizes graph neural networks (GNNs) and the graph information bottleneck (GIB) principle to create a compact, informative, and robust graph representation suitable for transmission. The challenge lies in the irregular structure of graph data, making GIB optimization complex. We address this by deriving a tractable variational upper bound for the objective function. Additionally, we propose the VQ-GIB mechanism, integrating vector quantization (VQ) to convert subgraph representations into a discrete codebook sequence, compatible with existing digital communication systems. Our experiments show that this GIB-based method significantly lowers communication costs while preserving essential task-related information. The approach demonstrates robust performance across various communication channels, suitable for both continuous and discrete systems.

Figures (7)

• Figure 1: Task-Oriented Communication Scheme for Graph Data: In this scheme, the “ transmitted codeword ” refers to the encoded representation of the task-related subgraph, while the “ corrupted codeword ” denotes the representation of the subgraph received by the receiver, which has undergone corruption during transmission through the channel. The symbol $\hat{Y}$ represents the task inference output.
• Figure 2: Training Framework for GIB-Based Task-Oriented Communication Systems: The process begins with training the mutual information estimator to acquire a set of optimized MINE parameters, denoted as ${{\kappa }^{*}}$. Following this, the entire neural network undergoes training, during which the mutual information estimator utilizes the previously determined MINE parameters.
• Figure 3: System Digitization Modules. This figure illustrates the process where the subgraph representation is matched against a codebook to generate an index sequence via the nearest neighbor principle. This sequence is sent through a symmetric discrete channel, subject to a specified error probability, leading to potential discrepancies between the transmitted and received index sequences. The receiver then reconstructs the representation vector using a shared codebook with the transmitter for subsequent task inference.
• Figure 4: Classification accuracy variation with SNR for three methods on PROTEINS and COLLAB Datasets using GCN backbone, at hidden dimensions of 16 and 32.
• Figure 5: Classification accuracy variation with SNR for the proposed method on PROTEINS and COLLAB datasets, utilizing GCN and GIN backbones at hidden dimensions of 16 and 32.