Graph Hopfield Networks: Energy-Based Node Classification with Associative Memory

Researchers introduce Graph Hopfield Networks, a new neural network architecture that combines associative memory with graph-based learning for node classification tasks. The method shows improvements of up to 5 percentage points on robustness tests and 2 percentage points on citation networks, outperforming standard baselines across multiple graph types.

• →Graph Hopfield Networks combine associative memory retrieval with graph Laplacian smoothing for enhanced node classification.
• →The method achieves up to 2 percentage points improvement on sparse citation networks and 5 percentage points additional robustness under feature masking.
• →The iterative energy-descent architecture serves as a strong inductive bias, with all variants outperforming standard baselines on Amazon co-purchase graphs.
• →The approach enables graph sharpening for heterophilous benchmarks without requiring architectural modifications.
• →Memory retrieval provides regime-dependent benefits, with effectiveness varying based on the specific graph structure and task.