Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

Graph convolutions that can finally model local structure

About

Despite quick progress in the last few years, recent studies have shown that modern graph neural networks can still fail at very simple tasks, like detecting small cycles. This hints at the fact that current networks fail to catch information about the local structure, which is problematic if the downstream task heavily relies on graph substructure analysis, as in the context of chemistry. We propose a very simple correction to the now standard GIN convolution that enables the network to detect small cycles with nearly no cost in terms of computation time and number of parameters. Tested on real life molecule property datasets, our model consistently improves performance on large multi-tasked datasets over all baselines, both globally and on a per-task setting.

R\'emy Brossard, Oriel Frigo, David Dehaene• 2020

Related benchmarks

TaskDatasetResultRank
Graph Classificationogbg-molpcba (test)
AP29.79
206
Quantum Chemical PredictionPCQM4M v2 (val)
MAE0.1167
68
Graph RegressionPeptides-struct
MAE0.3496
51
Graph property predictionPCQM4M-LSC (val)
MAE0.143
48
Molecular property predictionMOLPCBA OGB (test)
AP (Test)29.79
36
Graph ClassificationPeptides func
AP59.3
22
Graph property regressionPCQM4M (val)
MAE0.143
19
Graph property predictionPCQM4M-LSC (train)
MAE0.1248
16
Graph ClassificationOGBG-MOLPCBA (val)
AP30.65
15
Molecule ClassificationHIV scaffold (test)
ROC AUC0.788
10
Showing 10 of 18 rows

Other info

Follow for update