Abstract: This report synthesises findings from 12 peer-reviewed papers addressing the following research question: How does the performance (accuracy, recall) of graph contrastive learning models scale with increasing graph size in cross-domain recommendation tasks, as evaluated on heterogeneous datasets like. Contrastive…

Abstract: This report synthesises findings from 8 peer-reviewed papers addressing the following research question: Can hybrid approaches combining synthetic graph augmentation with contrastive learning objectives outperform pure variational inference baselines in few-shot node classification tasks on large-scale. Graphs are…

Abstract: This report synthesises findings from 9 peer-reviewed papers addressing the following research question: How does extreme graph sparsity affect the ranking accuracy (HR@K) of simple versus complex graph contrastive learning architectures when adapted for multimodal recommendation tasks. Contrastive learning (CL) has…

Abstract: This report synthesises findings from 8 peer-reviewed papers addressing the following research question: How does the robustness of minimalistic graph contrastive learning methods scale against adversarial node perturbations compared to complex augmentation pipelines in recommendation systems. Contrastive learning…

Abstract: This report synthesises findings from 11 peer-reviewed papers addressing the following research question: What is the impact of replacing graph augmentations with simple noise injection on the ranking accuracy of contrastive learning models across sparse user-item interaction datasets. Contrastive learning (CL) has…

Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: How does the robustness of LightGCL to noisy user-item interactions compare to SimGCL and DCL when evaluated using recall@k metrics on perturbed datasets. Human-Object Interaction (HOI) detection is crucial for…

Abstract: This report synthesises findings from 10 peer-reviewed papers addressing the following research question: How does the inference throughput of LightGCL compare to SimGCL and DCL on large-scale recommendation graphs when measured in samples per second. Graph neural network (GNN) is a powerful learning approach for…

Abstract: This report synthesises findings from 8 peer-reviewed papers addressing the following research question: What is the impact of varying graph sparsity levels on the convergence rate of LightGCL versus other contrastive learning methods in terms of training epochs required. Graph neural network (GNN) is a powerful…

Abstract: This report synthesises findings from 12 peer-reviewed papers addressing the following research question: How does the scaling of perturbation levels affect the inference efficiency of XSimGCL and other GCL methods on the Reddit dataset while maintaining a minimum NDCG@10 score of 0.7. Leading graph contrastive…

Abstract: This report synthesises findings from 11 peer-reviewed papers addressing the following research question: How does the adversarial robustness of XSimGCL compare to other GCL methods when evaluated on the Amazon dataset using NDCG@10 and NDCG@20 under similar perturbation levels. Contrastive learning (CL) has emerged…

Abstract: This report synthesises findings from 13 peer-reviewed papers addressing the following research question: What is the impact of using graph sparsification techniques on the inference efficiency (measured in FLOPs and latency) of contrastive learning-based graph clustering models when evaluated on. Contrastive…

Abstract: This report synthesises findings from 13 peer-reviewed papers addressing the following research question: How does the integration of multi-modal embeddings from both structural and attribute data improve the robustness of template-based graph clustering models against adversarial edge perturbations, as. Generating…

Abstract: This report synthesises findings from 13 peer-reviewed papers addressing the following research question: How do different contrastive learning augmentation strategies in graph neural networks perform in terms of node clustering accuracy (NMI) when applied to attributed graphs with varying sparsity. In order to…

Abstract: This report synthesises findings from 9 peer-reviewed papers addressing the following research question: What is the robustness of contrastive learning-based outlier detection methods in graph neural networks to adversarial attacks or noisy attributes, and how does this compare to traditional anomaly. Anomaly…

Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: To what extent does integrating adversarial training into GCN-based code generators improve cross-language generalization under diffusion-based attack scenarios. Real-time traffic prediction models play a pivotal…

Abstract: This report synthesises findings from 10 peer-reviewed papers addressing the following research question: How does adversarial training affect the F1-score robustness of Graph Code Generation models against diffusion-based perturbations compared to standard training. Generating high-fidelity and biologically…

Abstract: This report synthesises findings from 7 peer-reviewed papers addressing the following research question: What is the impact of different multi-view aggregation strategies (e.g., attention-based vs. max-pooling) on the detection F1-scores across diverse cross-domain graph datasets, as benchmarked on. Anomaly detection…

Abstract: This report synthesises findings from 12 peer-reviewed papers addressing the following research question: How does POEM's robustness to view dropout rates compare to state-of-the-art multi-view GANs when evaluated on the FreiHAND dataset under different occlusion conditions, measured by metrics like F1. Anomaly…

Abstract: This report synthesises findings from 9 peer-reviewed papers addressing the following research question: What is the impact of varying the number of basis points in POEM on the reconstruction accuracy in multi-view hand mesh tasks, as measured by metrics like mean per-vertex error on the HO-3D dataset. This work…

Abstract: This report synthesises findings from 6 peer-reviewed papers addressing the following research question: How does the integration of a point-embedded transformer in multi-view hand mesh reconstruction compare to traditional attention mechanisms in terms of inference efficiency on benchmark datasets like. This work…

Abstract: This report synthesises findings from 8 peer-reviewed papers addressing the following research question: Do multi-view graph anomaly detection frameworks with view dropout mechanisms achieve higher inference efficiency and accuracy trade-offs than single-view methods on large-scale heterogeneous graphs. Anomaly…

Abstract: This report synthesises findings from 7 peer-reviewed papers addressing the following research question: What is the effect of aggregating multi-view information versus single-view representations on the robustness of semi-supervised graph anomaly detection models across diverse dataset domains. Anomaly detection is…

Abstract: This report synthesises findings from 9 peer-reviewed papers addressing the following research question: How does view dropout in multi-view graph neural networks impact cross-domain generalization accuracy compared to single-view baselines on heterogeneous graph anomaly detection benchmarks. Anomaly detection is…

Abstract: This report synthesises findings from 20 peer-reviewed papers addressing the following research question: Do Vision GNNs trained with gradient checkpointing exhibit improved generalization across different graph sizes and domains (e.g., social networks vs. molecular graphs) compared to smaller baselines,. Recent…

Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: How does gradient checkpointing impact the accuracy-throughput tradeoff when training Vision GNNs on multimodal datasets compared to traditional memory-efficient methods like gradient accumulation. Graph Neural…