RECORD
Reliable Representations Make A Stronger Defender: Unsupervised Structure Refinement for Robust GNN
Machine Learning · Submitted 30 Jun 2022
abstract compiled · 2026-09-23 · no language-model review
The authors' main claim Extensive experiments on four real-world graph benchmarks demonstrate that STABLE outperforms the state-of-the-art methods and successfully defends against various attacks.
The compiled record
The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.
Open the compiled record
The abstract, sentence by sentence
Benefiting from the message passing mechanism, Graph Neural Networks (GNNs) have been successful on flourish tasks over graph data. However, recent studies have shown that attackers can catastrophically degrade the performance of GNNs by maliciously modifying the graph structure. A straightforward solution to remedy this issue is to model the edge weights by learning a metric function between pairwise representations of two end nodes, which attempts to assign low weights to adversarial edges. The existing methods use either raw features or representations learned by supervised GNNs to model the edge weights. However, both strategies are faced with some immediate problems: raw features cannot represent various properties of nodes (e.g., structure information), and representations learned by supervised GNN may suffer from the poor performance of the classifier on the poisoned graph. We need representations that carry both feature information and as mush correct structure information as possible and are insensitive to structural perturbations. To this end, we propose an unsupervised pipeline, named STABLE, to optimize the graph structure. Finally, we input the well-refined graph into a downstream classifier. For this part, we design an advanced GCN that significantly enhances the robustness of vanilla GCN without increasing the time complexity. Extensive experiments on four real-world graph benchmarks demonstrate that STABLE outperforms the state-of-the-art methods and successfully defends against various attacks.
background contribution method result interpretation main claim
| # | sentence | role | the authors' own work | states the main conclusion | strength |
|---|---|---|---|---|---|
| s0 | Benefiting from the message passing mechanism, Graph Neural Networks (GNNs) have been successful on flourish tasks over graph data. | background 1.00 | 0.07 | 0.02 | asserted 1.00 |
| s1 | However, recent studies have shown that attackers can catastrophically degrade the performance of GNNs by maliciously modifying the graph structure. | background 0.99 | 0.04 | 0.02 | asserted 0.96 |
| s2 | A straightforward solution to remedy this issue is to model the edge weights by learning a metric function between pairwise representations of two end nodes, which attempts to assign low weights to adversarial edges. | background 0.78 | 0.06 | 0.04 | asserted 0.89 |
| s3 | The existing methods use either raw features or representations learned by supervised GNNs to model the edge weights. | background 1.00 | 0.05 | 0.02 | asserted 1.00 |
| s4 | However, both strategies are faced with some immediate problems: raw features cannot represent various properties of nodes (e.g., structure information), and representations learned by supervised GNN may suffer from the poor performance of the classifier on the poisoned graph. | gap 0.73 | 0.10 | 0.03 | asserted 0.75 |
| s5 | We need representations that carry both feature information and as mush correct structure information as possible and are insensitive to structural perturbations. | gap 0.97 | 0.08 | 0.07 | asserted 0.84 |
| s6 | To this end, we propose an unsupervised pipeline, named STABLE, to optimize the graph structure. | contribution 0.83 | 0.90 | 0.07 | asserted 0.96 |
| s7 | Finally, we input the well-refined graph into a downstream classifier. | method 0.99 | 0.80 | 0.03 | asserted 0.85 |
| s8 | For this part, we design an advanced GCN that significantly enhances the robustness of vanilla GCN without increasing the time complexity. | method 0.87 | 0.90 | 0.05 | asserted 1.00 |
| s9 | Extensive experiments on four real-world graph benchmarks demonstrate that STABLE outperforms the state-of-the-art methods and successfully defends against various attacks. | result 1.00 | 0.96 | 0.77 | asserted 1.00 |
What was examined, and what was not
- arXiv listing abstract v4
- full text
- figures and tables
- references and citing sentences
- relations to other papers
- P abstract
- S sections
- T body sentences
- C citations
- R cross-paper relations
The paper
Original abstract
Benefiting from the message passing mechanism, Graph Neural Networks (GNNs) have been successful on flourish tasks over graph data. However, recent studies have shown that attackers can catastrophically degrade the performance of GNNs by maliciously modifying the graph structure. A straightforward solution to remedy this issue is to model the edge weights by learning a metric function between pairwise representations of two end nodes, which attempts to assign low weights to adversarial edges. The existing methods use either raw features or representations learned by supervised GNNs to model the edge weights. However, both strategies are faced with some immediate problems: raw features cannot represent various properties of nodes (e.g., structure information), and representations learned by supervised GNN may suffer from the poor performance of the classifier on the poisoned graph. We need representations that carry both feature information and as mush correct structure information as possible and are insensitive to structural perturbations. To this end, we propose an unsupervised pipeline, named STABLE, to optimize the graph structure. Finally, we input the well-refined graph into a downstream classifier. For this part, we design an advanced GCN that significantly enhances the robustness of vanilla GCN without increasing the time complexity. Extensive experiments on four real-world graph benchmarks demonstrate that STABLE outperforms the state-of-the-art methods and successfully defends against various attacks.
Human review
How the literature uses it
1 arXiv paper cite this in 2 sentences
Papers citing it · 1 sentence without a confident label
Cites it as background 1 paper
-
A number of methods have been proposed to mitigate specific data-quality problems, including text denoising and completion [ref] , graph structure learning [ref] , label correction, imbalance-aware training [ref] , and LLM-assisted graph modeling [ref] .
OpenRTAG: A Comprehensive Benchmark for Robust Text-Attributed Graph Learning under Data Quality Degradation · Introduction · 0.83
What it cites · 10 sentences without a confident label
Cites as background 5 papers
-
[-15]In the methods mentioned above, one representative type of approach is graph structure learning [ref] , which aims to detect the potential adversarial edges and assign these edges lower weights or even remove them.
A Survey on Graph Structure Learning: Progress and Opportunities · Robust GNNs · 0.97
-
Recently, GRACE [ref] and GraphCL [ref] adapted the data augmentation methods in vision [ref] to graphs and achieved state-of-the-art performance.
Deep Graph Contrastive Representation Learning · Graph Contrastive Learning · 0.84
-
To learn task-irrelevant representations, we consider using unsupervised methods, which are mainly divided into three categories: GAEs [ref] , random walk methods [ref] , and contrastive methods [ref] .
Deep Graph Infomax · Graph Contrastive Learning · 0.99
Discussion
Know something about this paper? Leave a comment, or highlight any passage above to comment on it.