Abstract: This report synthesises findings from 11 peer-reviewed papers addressing the following research question: How do different graph augmentation strategies in GCAD methods impact the trade-off between anomaly detection accuracy and training time on large-scale graphs like ogbn-products. Federated learning (FL) is a…

Abstract: This report synthesises findings from 4 peer-reviewed papers addressing the following research question: What is the effect of topology-preserving versus feature-masking augmentations on the F1-score of self-supervised graph anomaly detectors across varying graph densities. Abstract Deep learning (DL) is…

Abstract: This report synthesises findings from 16 peer-reviewed papers addressing the following research question: How do contrastive graph augmentation strategies impact the AUC-ROC performance of self-supervised GCAD models compared to supervised baselines on node-level anomaly detection. Deep models trained in supervised…

Abstract: This report synthesises findings from 9 peer-reviewed papers addressing the following research question: Can the feature imputation mechanism of AmGCL be adapted to enhance multimodal graph representation learning where node attributes are derived from heterogeneous data sources. Single-cell RNA-sequencing…

Abstract: This report synthesises findings from 11 peer-reviewed papers addressing the following research question: To what extent does the self-supervised contrastive learning strategy in AmGCL improve robustness against high-percentage attribute missingness compared to standard graph imputation baselines. Graph Neural…

Abstract: This report synthesises findings from 13 peer-reviewed papers addressing the following research question: How do graph contrastive anomaly detection models scale with graph size compared to supervised methods when evaluated on heterophilic graphs using the F1 score as the primary metric. With a long history of…

Abstract: This report synthesises findings from 4 peer-reviewed papers addressing the following research question: What is the robustness of graph contrastive anomaly detection models against adversarial perturbations compared to supervised methods when measured by detection accuracy on perturbed Amazon co-author. Machine…

Abstract: This report synthesises findings from 8 peer-reviewed papers addressing the following research question: How robust is Mul-GAD's anomaly detection performance across different domains of heterophilic graphs (e.g., social networks vs. citation networks) when evaluated using consistent benchmark datasets. Abstract Data…

Abstract: This report synthesises findings from 8 peer-reviewed papers addressing the following research question: How does the inference latency of graph contrastive anomaly detection models compare to supervised GNN baselines when evaluated on the ogbn-arxiv benchmark using throughput (queries per second) as. Systems for…

Abstract: This report synthesises findings from 6 peer-reviewed papers addressing the following research question: How does the performance of Mul-GAD compare to spatial GNN baselines on heterophilic graph benchmark datasets in terms of AUC-ROC and F1-score metrics. With a long history of traditional Graph Anomaly Detection…

Abstract: This report synthesises findings from 1 peer-reviewed paper addressing the following research question: What is the computational efficiency trade-off of Mul-GAD versus GNN baselines when scaling to large-scale heterophilic graphs, measured in terms of inference time and memory usage. \<p\>Transfer learning, as…

Abstract: This report synthesises findings from 10 peer-reviewed papers addressing the following research question: How does the sample efficiency of Mul-GAD in low-label regimes compare to recent contrastive learning approaches for graph anomaly detection. Anomaly detection has been used for decades to identify and extract…

Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: To what extent does Mul-GAD maintain detection accuracy under adversarial feature perturbations compared to state-of-the-art semi-supervised graph anomaly detection frameworks. Abstract Deep learning (DL) is…

Abstract: This report synthesises findings from 1 peer-reviewed paper addressing the following research question: How does the inference throughput of Mul-GAD compare to spectral-based GNN anomaly detectors when scaling to graphs with over 100,000 nodes. \<p\>Transfer learning, as a key paradigm in modern machine…

Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: Does replacing concatenation with attention-based fusion in multimodal alignment frameworks improve sample efficiency and downstream task performance on zero-shot classification benchmarks. Today, despite decades…

Abstract: This report synthesises findings from 12 peer-reviewed papers addressing the following research question: How do attention-based multi-view fusion mechanisms compare to concatenation strategies in multimodal large language models regarding inference latency and accuracy on standard VQA benchmarks. Precision and…

Abstract: This report synthesises findings from 7 peer-reviewed papers addressing the following research question: What is the impact of contrastive alignment strategies on the inference efficiency and reasoning accuracy of code generation models when evaluated on the HumanEval benchmark with obfuscated inputs. Large language…

Abstract: This report synthesises findings from 13 peer-reviewed papers addressing the following research question: To what extent does semi-supervised contrastive learning improve cross-domain generalization for anomaly detection in attributed networks compared to unsupervised approaches when facing structural. A…

Abstract: This report synthesises findings from 15 peer-reviewed papers addressing the following research question: To what extent does multi-scale contrastive pre-training improve the robustness of large language models against adversarial text perturbations as measured by accuracy drop on GLUE benchmark tasks.…

Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: How does the semi-supervised multi-view aggregation approach in Mul-GAD compare to fully unsupervised GNN-based methods in terms of AUC performance on heterophilic graphs under adversarial edge. Benefiting from…

Abstract: This report synthesises findings from 11 peer-reviewed papers addressing the following research question: What is the computational efficiency trade-off of Mul-GAD's multi-view aggregation mechanism compared to single-view GNN-based anomaly detection methods when scaling to large heterophilic graphs. Recent years…

Abstract: This report synthesises findings from 4 peer-reviewed papers addressing the following research question: How does Mul-GAD's performance on heterophilic graph anomaly detection compare to recent state-of-the-art semi-supervised GNN methods when evaluated on benchmark datasets like OGB and TuSimple using. Convolutional…

Abstract: This report synthesises findings from 11 peer-reviewed papers addressing the following research question: How do contrastive self-supervised learning objectives impact the reasoning robustness of graph neural networks against adversarial neighbor distribution shifts compared to standard autoencoder. Point clouds…

Abstract: This report synthesises findings from 6 peer-reviewed papers addressing the following research question: How does the performance of Mul-GAD in semi-supervised graph anomaly detection compare to other state-of-the-art GNN-based methods (e.g., DOMINANT, GraphGAN) on benchmark datasets like Cora, PubMed,. With a long…

Abstract: This report synthesises findings from 10 peer-reviewed papers addressing the following research question: What is the efficiency trade-off between Mul-GAD and alternative semi-supervised graph anomaly detection methods in terms of inference latency and memory usage when scaled to large graphs (e.g.,. Anomaly…