Research map / Graph learning / GNN architectures

Let Your Features Tell The Differences: Understanding Graph Convolution By Feature Splitting

ICLR 2025 · Graph learning › GNN architectures · approach: Simple and scalable GNNs

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Investigates whether graph convolution benefits every feature dimension equally, and proposes a feature-splitting approach that treats features differently during graph convolution.

Problem

Traditional GNNs apply graph convolution uniformly to all feature dimensions, assuming each is equally beneficial; the paper questions whether convolution can harm some feature dimensions.

Approach

Splits feature dimensions and treats them differently under graph convolution, so that convolution is applied only where it is beneficial rather than uniformly across all features.

Surprising findings

• Graph convolution may have harmful effects on certain feature dimensions, not uniformly beneficial.

Experimental setup

Builds on: graph convolutional network

Closest papers by meaning

• Feature Distribution on Graph Topology Mediates the Effect of Graph Convolution: Homophily Perspective (ICML 2024 · similarity 0.90)
• Can Classic GNNs Be Strong Baselines for Graph-level Tasks? Simple Architectures Meet Excellence (ICML 2025 · similarity 0.89)
• Classic GNNs are Strong Baselines: Reassessing GNNs for Node Classification (NeurIPS 2024 · similarity 0.89)
• Diffusing to the Top: Boost Graph Neural Networks with Minimal Hyperparameter Tuning (ICLR 2025 · similarity 0.88)
• A Manifold Perspective on the Statistical Generalization of Graph Neural Networks (ICML 2025 · similarity 0.88)
• On the Stability of Graph Convolutional Neural Networks: A Probabilistic Perspective (NeurIPS 2025 · similarity 0.88)