Back to feed
arXiv cs.LG
arXiv cs.LG
5/13/2026
Hierarchical Multi-Scale Graph Neural Networks: Scalable Heterophilous Learning with Oversmoothing and Oversquashing Mitigation

Hierarchical Multi-Scale Graph Neural Networks: Scalable Heterophilous Learning with Oversmoothing and Oversquashing Mitigation

Short summary

Introduces HMH (Hierarchical Multi-view HAAR), a novel spectral Graph Neural Network framework for heterophilous graphs where adjacent nodes carry different labels. Uses feature-aware signed affinities and hierarchical Haar basis filters to prevent hub-dominated aggregation and signal bottlenecking while maintaining near-linear time complexity. Demonstrates 3–7% improvements in node and graph classification accuracy on standard benchmarks.

  • •Novel spectral GNN framework addressing heterophily, oversmoothing, and oversquashing problems
  • •Constructs sparse, orthonormal Haar basis at each hierarchical level for learnable spectral filtering
  • •Achieves 3–7% accuracy gains over state-of-the-art spectral baselines with linear scalability

Generated with AI, which can make mistakes.

#research-breakthrough
Read full article at arXiv cs.LG

Is this a good recommendation for you?

Explore more

Path-Based Gradient Boosting for Graph-Level Prediction
arXiv cs.LG
Structural Interpretations of Protein Language Model Representations via Differentiable Graph Partitioning
arXiv cs.LG
Belief or Circuitry? Causal Evidence for In-Context Graph Learning
arXiv CS.AI
TTCD:Transformer Integrated Temporal Causal Discovery from Non-Stationary Time Series Data
arXiv cs.LG