GABO: Graph Augmentations with Bi-level Optimization

TL;DR

This paper introduces GABO, a bilevel optimization-based data augmentation method for graph classification, achieving state-of-the-art results on the ogbg-molhiv dataset with a GIN+virtual classifier.

Contribution

The paper presents a novel graph augmentation framework using bilevel optimization that outperforms existing augmentation techniques like FLAG.

Findings

Achieved a test ROCAUC of 77.77% on ogbg-molhiv.

Outperforms state-of-the-art FLAG augmentation with the same classifier.

Demonstrates effectiveness of bilevel optimization for graph data augmentation.

Abstract

Data augmentation refers to a wide range of techniques for improving model generalization by augmenting training examples. Oftentimes such methods require domain knowledge about the dataset at hand, spawning a plethora of recent literature surrounding automated techniques for data augmentation. In this work we apply one such method, bilevel optimization, to tackle the problem of graph classification on the ogbg-molhiv dataset. Our best performing augmentation achieved a test ROCAUC score of 77.77 % with a GIN+virtual classifier, which makes it the most effective augmenter for this classifier on the leaderboard. This framework combines a GIN layer augmentation generator with a bias transformation and outperforms the same classifier augmented using the state-of-the-art FLAG augmentation.