Abstract: This report synthesises findings from 9 peer-reviewed papers addressing the following research question: How does the choice of contrastive loss function in AmGCL affect the AUC-ROC for anomaly detection in graphs with varying percentages of missing node attributes. Attribute graphs are ubiquitous in multimedia…

Abstract: This report synthesises findings from 9 peer-reviewed papers addressing the following research question: How does the inference efficiency of graph contrastive anomaly detection models compare to supervised methods when evaluated on large-scale heterophilic graph benchmarks like Amazon, Coauthor, or. Combining Graph…

Abstract: This report synthesises findings from 7 peer-reviewed papers addressing the following research question: How does the robustness of Mul-GAD's anomaly detection performance vary across different types of heterophilic graphs (e.g., social networks, citation networks) compared to spatial GNN baselines. Anomaly detection…

Abstract: This report synthesises findings from 11 peer-reviewed papers addressing the following research question: How does the F1-score performance of Mul-GAD compare to state-of-the-art spectral-based graph anomaly detection methods on heterophilic graphs under feature masking rates of 30\%, 50\%, and 70\%. Anomaly…

Abstract: This report synthesises findings from 13 peer-reviewed papers addressing the following research question: What is the computational efficiency trade-off between Mul-GAD and spatial GNN baselines when scaling to large heterophilic graphs with varying sparsity levels. Heterogeneous graph neural networks (HGNNs) were…

Abstract: This report synthesises findings from 8 peer-reviewed papers addressing the following research question: What is the impact of multi-view fusion strategies (e.g., attention-based vs. concatenation-based) in the Mul-GAD framework on downstream task performance, as evaluated through both accuracy and. Anomaly detection…

Abstract: This report synthesises findings from 8 peer-reviewed papers addressing the following research question: To what extent does incorporating contrastive learning (e.g., via GraphCL) into the Mul-GAD framework improve robustness against adversarial attacks in cross-domain graph anomaly detection, as. Combining Graph…

Abstract: This report synthesises findings from 8 peer-reviewed papers addressing the following research question: How do different GNN architectures (e.g., GraphSAGE vs. GAT) compare in terms of inference latency and memory efficiency when deployed in the Mul-GAD framework for large-scale graph anomaly detection. Anomaly…

Abstract: This report synthesises findings from 15 peer-reviewed papers addressing the following research question: What is the impact of neighbor distribution shifts caused by adversarial attacks on the robustness of GNN-based anomaly detectors in semi-supervised versus unsupervised settings. Anomaly detection on attributed…

Abstract: This report synthesises findings from 11 peer-reviewed papers addressing the following research question: How does semi-supervised graph anomaly detection performance on heterophilic graphs compare to fully unsupervised methods when evaluated under adversarial structural perturbations using AUC metrics. Anomaly…

Abstract: This report synthesises findings from 7 peer-reviewed papers addressing the following research question: How does the robustness of Mul-GAD compare to other semi-supervised graph anomaly detection methods (e.g., DOMINANT or GAN-based approaches) when evaluated on perturbed graph structures with varying. Anomaly…

Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: How does the scalability of Mul-GAD compare to other GNN-based approaches such as GraphSAGE or GAT in terms of inference time and memory consumption when evaluated on large-scale graph datasets like. Anomaly…

Abstract: This report synthesises findings from 10 peer-reviewed papers addressing the following research question: What is the performance gap between Mul-GAD and traditional shallow learning methods (e.g., SVM, Random Forest) on benchmark datasets like CORA or Citeseer when measured by F1-score and AUC-ROC. Sentiment…

Abstract: This report synthesises findings from 16 peer-reviewed papers addressing the following research question: What is the impact of Directional Preference Alignment versus standard RLHF on the syntactic robustness of Code LLMs when evaluated against adversarial syntax perturbations in the HumanEval dataset. Fine-grained…

Abstract: This report synthesises findings from 13 peer-reviewed papers addressing the following research question: How does Directional Preference Alignment compare to RLHF in improving code generation accuracy for low-resource programming languages on the MultiPL-E benchmark. Fine-grained control over large language models…

Abstract: This report synthesises findings from 6 peer-reviewed papers addressing the following research question: How does asynchronous federated learning with latency-aware aggregation impact the convergence accuracy of multimodal malware detection models compared to synchronous baselines. This work investigates the…

Abstract: This report synthesises findings from 15 peer-reviewed papers addressing the following research question: What is the effect of varying the fraction of clients participating in each communication round on the inference efficiency and zero-shot accuracy of federated multimodal models on the VQA-v2. A significant…

Abstract: This report synthesises findings from 6 peer-reviewed papers addressing the following research question: How does the use of adaptive client sampling based on update norm magnitude in federated learning impact the convergence speed and accuracy of multimodal vision-language models on the VQA-v2. It is well understood…

Abstract: This report synthesises findings from 11 peer-reviewed papers addressing the following research question: What is the optimal model scaling factor for code generation quality (measured by HumanEval pass@10) when training federated LLMs with partial client participation and IID versus non-IID data. The proliferation…

Abstract: This report synthesises findings from 10 peer-reviewed papers addressing the following research question: How does the scaling of federated clients with non-IID data affect the convergence speed and final accuracy of multimodal models, as evaluated on cross-modal retrieval benchmarks like Flickr30K or. Federated…

Abstract: This report synthesises findings from 15 peer-reviewed papers addressing the following research question: Does FedDiverse's approach to handling non-IID data improve the robustness of CodeLlama against domain-shifted code benchmarks compared to standard federated averaging. Federated Learning (FL) enables…

Abstract: This report synthesises findings from 10 peer-reviewed papers addressing the following research question: How does diversity-driven client selection in federated learning impact the pass@1 code generation scores of CodeLlama on HumanEval under high statistical heterogeneity. Federated Learning (FL) enables…

Abstract: This report synthesises findings from 10 peer-reviewed papers addressing the following research question: What is the impact of varying degrees of non-IID data on the alignment efficiency of collaborative multimodal models in federated learning, as measured by structural similarity metrics (e.g., cosine. Federated…

Abstract: This report synthesises findings from 10 peer-reviewed papers addressing the following research question: How does dynamic meta-layer aggregation perform in terms of inference efficiency and communication overhead compared to federated averaging when defending against Byzantine attacks on MNIST and. We propose…

Abstract: This report synthesises findings from 9 peer-reviewed papers addressing the following research question: How do FLCert's defense mechanisms affect the convergence speed and communication efficiency in federated learning settings with heterogeneous client capabilities, measured on standard FL benchmarks. Federated…