Paper Overview
Field: Machine Learning Author: Fabricio Breve Published: 2026-09-18 arXiv: 2609.22053
Abstract
Graph Convolutional Networks (GCNs) are highly sensitive to label noise, since corrupted supervision can propagate through the graph and degrade learned node representations. This work proposes PCC+GCN, a hybrid framework that uses Particle Competition and Cooperation (PCC) as a graph-based label-refinement stage before GCN training.
PCC identifies suspicious labeled nodes through particle domination dynamics and determines whether their labels should be preserved, removed, or reassigned before GCN training. The framework also allows the graph used by PCC to be augmented with feature-based \(k\)-nearest-neighbor edges, while the GCN itself is trained on the original graph structure and node features.
Key Results
- Benchmark: Evaluated on ten graph datasets from the NoisyGL benchmark, covering conventional Uniform, Pair, and Random label noise, as well as instance-dependent label noise. A detailed hyperparameter analysis was conducted on Cora, CiteSeer, and PubMed.
- Conventional noise: PCC+GCN achieved the highest overall average accuracy and the best average rank among all evaluated methods, outperforming baseline GCN by 1.67 percentage points on average in the clean setting and across all noise scenarios.
- Instance-dependent noise: PCC+GCN remained competitive with the most robust methods while requiring significantly less execution time — it was the fastest robust method on eight of the ten datasets.
- Conclusion: PCC-based label refinement is an effective and computationally efficient preprocessing strategy for improving GCN robustness under noisy supervision.
Link
Full paper: https://arxiv.org/abs/2609.22053