Structure-Feature Aligned Graph Learning via Alternating Constrained Optimization

2026-07-13

AI Summary

This work addresses the performance degradation of conventional graph neural networks (GNNs) caused by the tight coupling of feature transformation and neighborhood aggregation, which renders them sensitive to topological noise and heterophilic connections. To mitigate this issue, the authors propose a dual-view constraint framework that effectively decouples feature and topological dependencies by alternately aligning structure-aware GNN embeddings with feature priors learned from a structure-agnostic anchor model. The approach introduces an independent anchor network to capture intrinsic node attributes and designs a Channel-Split Adaptive Gating GNN (CSAG-GNN) that dynamically fuses global spectral smoothness with local spatial discriminative information. Combined with node-level gating and a stable alternating optimization strategy, the method suppresses representation drift. Extensive experiments demonstrate consistently improved performance across both homophilic and heterophilic benchmark datasets, significantly enhancing the model’s structural robustness.

Abstract

We introduce a constrained two-view framework for node prediction that aligns structure-conditioned GNN embeddings with a structure-free feature prior learned by an anchor model. Conventional Graph Neural Networks (GNNs) couple feature transformation and neighborhood aggregation, which renders them vulnerable to topology noise and heterophilous connections. To decouple this dependency, our framework utilizes an independent anchor network to capture intrinsic attribute features via a self-supervised reconstruction objective. Furthermore, we propose a Channel-Split Adaptive Gated GNN (CSAG-GNN) that dynamically routes representations between global spectral smoothing and local spatial discrimination through a node-wise gating mechanism. We propose a stable cyclic alternating optimization strategy to solve the resulting coupled bi-level objective, preventing mutual representation drift during training. Empirical results on both homophilous and heterophilous benchmarks show balanced performance gains and structural robustness over competitive baselines.

Problem

• Graph Neural Networks
• topology noise
• heterophilous connections
• feature-structure coupling
• node prediction

Innovation

• Structure-Feature Alignment
• Alternating Constrained Optimization
• Anchor Model
• CSAG-GNN
• Heterophilous Graphs