Research map / Graph learning / GNN architectures

CKGConv: General Graph Convolution with Continuous Kernels

ICML 2024 · Graph learning › GNN architectures · approach: Deep graph convolution and simplification · 1 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

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

Proposes CKGConv, a general graph convolution that parameterizes the kernel as a continuous function of node-pair geometry, unifying spatial and spectral graph convolutions.

Problem

Existing graph convolutions (spatial or spectral) are inflexible and not unified, and defining a general convolution on graphs is hard due to lack of canonical coordinates, irregular structure, and graph symmetries.

Approach

Parameterizes the convolution kernel as a continuous function over pairwise graph geometry (continuous kernel on graph coordinates), so that a single operator generalizes multiple existing spatial and spectral graph convolutions.

Experimental setup

Baselines: GCN GAT GIN ChebNet

Builds on: GCN message passing neural networks

Closest papers by meaning

• Convolutional Kernel Networks for Graph-Structured Data (ICML 2020 · similarity 0.91)
• Simple Spectral Graph Convolution (ICLR 2021 · similarity 0.91)
• All in a Row: Compressed Convolution Networks for Graphs (ICML 2023 · similarity 0.90)
• FastGCN: Fast Learning with Graph Convolutional Networks via Importance Sampling (ICLR 2018 · similarity 0.90)
• Deeper Insights Into Graph Convolutional Networks for Semi-Supervised Learning (AAAI 2018 · similarity 0.90)
• Generalizing CNNs to graphs with learnable neighborhood quantization (NeurIPS 2024 · similarity 0.89)