AI Paper Atlas

Research map / Graph learning / GNN architectures

Orthogonal Bases for Equivariant Graph Learning with Provable k-WL Expressive Power

Jia He, Maggie Cheng

JMLR 2025 · Graph learning › GNN architectures · approach: Subgraph and distance-aware GNNs

Read the paperCheck your idea against 137,217 papers

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
use order-3 GNN-a instead of PPGNOrder-3 GNN-a · TUDatasets training time (PROTEINS)training time per epoch (s): 865 -> 223Improved
use order-3 GNN-a instead of PPGNOrder-3 GNN-a · MUTAG parameters in linear layersparameters: 2409600 -> 196500Improved
use order-2 GNN-b instead of PPGNOrder-2 GNN-b · QM9training time per epoch (s): 3300 -> 70Improved
use order-3 GNN-a instead of PPGNOrder-3 GNN-a · QM9training time per epoch (s): 3300 -> 178Improved
use order-3 GNN-a instead of order-2 GNN-bOrder-3 GNN-a · TUDatasets NCI1accuracy (%): 80 -> 85Improved
use order-3 GNN-a instead of order-2 GNN-bOrder-3 GNN-a · TUDatasets PROTEINSaccuracy (%): 73.9 -> 77.8Improved
use order-3 GNN-a instead of order-2 GNN-bOrder-3 GNN-a · TUDatasets PTCaccuracy (%): 72.4 -> 72.1No clear effect
use order-3 GNN-a instead of order-2 GNN-bOrder-3 GNN-a · ZINC12kMAE: 0.146 -> 0.110Improved
use order-3 GNN-a instead of order-2 GNN-bOrder-3 GNN-a · QM9 (Δϵ)MAE: 0.0025 -> 0.0033Hurt
use order-2 GNN-b instead of PPGNOrder-2 GNN-b · ZINC12kMAE: 0.256 -> 0.146Improved
use order-2 GNN-b (linear layers) instead of 2-WLOrder-2 GNN-b · graph isomorphism tests 1-2succeeded: N -> YImproved
use order-2 GNN-a instead of order-2 GNN-b (ablation)Order-2 GNN-a · graph isomorphism tests 1-2succeeded: Y -> NHurt

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