AI Paper Atlas

Research map / Graph learning / Graph self-supervised learning

Structure-free Graph Condensation: From Large-scale Graphs to Condensed Graph-free Data

Xin Zheng, Miao Zhang, Chunyang Chen, Quoc Viet Hung Nguyen, Xingquan Zhu, Shirui Pan

NeurIPS 2023 · Graph learning › Graph self-supervised learning · approach: Heterophily and imbalance on graphs · 9 citations (OpenAlex)

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
drop learned graph structure (use identity matrix, structure-free) versus synthesizing kNN, cosine or parameterized structures (ablation)SFGC · Cora/Citeseer/Ogbn-arxiv, nine ratiosaccuracyImproved
use parameterized graph structure learning module in condensation (ablation)SFGC · transductive datasetsaccuracyHurt
use continuous cosine-similarity structure instead of discrete kNN structure (ablation)SFGC · transductive datasetsaccuracyHurt
choose kNN k per dataset and ratio (k=1,2,5) (ablation)SFGC-d variants · CiteseeraccuracyMixed
evaluate condensed data with closed-form GNTK score rather than iterative GNN training (ablation)SFGC · Flickr r=0.1%evaluation time (s): 0.845s -> 0.015sImproved
evaluate condensed data with closed-form GNTK score rather than iterative GNN training (ablation)SFGC · Ogbn-arxiv r=0.05%evaluation time (s): 4.264s -> 0.042sImproved
structure-free trajectory meta-matching instead of GCOND online gradient matching with learned structuregraph condensation · Ogbn-arxiv r=0.05%accuracy (%): 59.2 -> 65.5Improved
SFGC versus GCOND, condensed on GCN and tested on other architecturesSFGC · Citeseer r=1.8%, GATaccuracy (%): 55.4 -> 72.1Improved
SFGC versus GCOND on Flickr and Reddit-style inductive settingsSFGC · Flickr r=0.5%accuracy (%): 47.1 -> 47.0No clear effect
SFGC versus GCOND across seven GNN architectures on FlickrSFGC · Flickr r=0.5% averageaccuracy (%): 45.6 -> 45.0Hurt

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