AI Paper Atlas

Research map / Graph learning / GNN architectures

Graph Neural Networks with Learnable and Optimal Polynomial Bases

Yuhe Guo, Zhewei Wei

ICML 2023 · Graph learning › GNN architectures · approach: Theory of graph convolutional networks · 8 citations (OpenAlex)

Read the paper

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
learn an optimal signal-specific polynomial basisOptBasisGNN · YCbCr synthetic filter learningMSE loss: 0.0058 vs 0.1501 (ChebII)Improved
learn an arbitrary orthonormal basis rather than the optimal one (ablation)polynomial filter learning · YCbCr synthetic filter learningMSE loss: 0.0058 vs 0.3175Hurt
use a Monomial basis rather than the optimal basispolynomial filter learning · YCbCr synthetic filter learningMSE loss: 0.0058 vs 3.9076Hurt
apply the scaled-up OptBasisGNN with batch trainingOptBasisGNN · Wikiaccuracy: 60.95 -> 61.85Improved
use learnable orthonormal polynomial basesFavardGNN · Squirrelaccuracy: 57.72 -> 63.49Improved
use the optimal polynomial basisOptBasisGNN · Chameleonaccuracy: 74.20 -> 74.26Improved
use the optimal polynomial basisOptBasisGNN · ogbn-arxivaccuracy: 72.32 -> 72.27No clear effect
train FavardGNN for extended epochs without early stopping (ablation)FavardGNN · YCbCr synthetic filter learningMSE lossHurt
use early stopping rather than continuing FavardGNN training (ablation)FavardGNN · node classification and filter learninglossImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy