Stuart-Landau Oscillatory Graph Neural Network

Deep graph neural networks can lose discriminative node information through oversmoothing and suffer from vanishing gradients. Oscillatory graph networks counter these effects with dynamical-system-inspired propagation, but earlier approaches largely focus on harmonic or phase-only dynamics.

The Complex-Valued Stuart–Landau Graph Neural Network (SLGNN) evolves both feature amplitudes and phases using coupled Stuart–Landau oscillators. Its Hopf and coupling parameters provide explicit control over amplitude regulation, network interaction, and synchronization behavior.

Across node classification, graph classification, and graph regression experiments, SLGNN outperforms the evaluated oscillatory graph neural network baselines, supporting amplitude-aware oscillatory dynamics as an expressive basis for deep graph architectures.

Paper: https://arxiv.org/abs/2511.08094