Abstract: This report synthesises findings from 9 peer-reviewed papers addressing the following research question: How do training loss trajectories and final accuracy differ between prototype-based and traditional embedding strategies in federated graph learning as the number of participating clients increases. Graph Neural…

Abstract: This report synthesises findings from 11 peer-reviewed papers addressing the following research question: What is the comparative inference efficiency of prototype-based embeddings against traditional embeddings in federated graph neural networks under varying client scalability on standard benchmarks. Graph Neural…

Abstract: This report synthesises findings from 16 peer-reviewed papers addressing the following research question: How do modern attention-based vision transformers compare to the five-layer convolutional architecture in terms of scaling efficiency and top-1 accuracy when trained on high-resolution image. Modern computer…

Abstract: This report synthesises findings from 13 peer-reviewed papers addressing the following research question: Does the use of dynamic prototype aggregation in federated graph learning improve robustness against non-IID data distributions on social network datasets like Reddit while reducing bandwidth. Graph Neural…

Abstract: This report synthesises findings from 11 peer-reviewed papers addressing the following research question: What is the impact of replacing max-pooling layers with strided convolutions on the memory usage and classification accuracy of 60-million parameter CNNs trained on large-scale image datasets. Convolutional…

Abstract: This report synthesises findings from 16 peer-reviewed papers addressing the following research question: Do graph contrastive pre-training methods improve few-shot node classification accuracy on heterogeneous information networks compared to traditional supervised fine-tuning under adversarial attacks. The…

Abstract: This report synthesises findings from 11 peer-reviewed papers addressing the following research question: What is the impact of varying label scarcity ratios on the convergence speed and final F1 performance of semi-supervised GNNs versus fully supervised counterparts on large-scale social network graphs. We…

Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: How does the inference throughput of deep convolutional neural networks on ImageNet vary when applying different quantization strategies while maintaining top-5 error rates below 17.0\%. Deep convolutional neural…

Abstract: This report synthesises findings from 13 peer-reviewed papers addressing the following research question: How does the robustness of multi-scale contrastive learning in Graph Neural Networks compare to supervised baselines on adversarially perturbed Reddit and Amazon datasets when evaluated using F1. Following the…

Abstract: This report synthesises findings from 11 peer-reviewed papers addressing the following research question: How does the adversarial robustness of graph convolutional networks compare to transformer-based architectures for code generation dependency graph imputation, as measured by BLEU scores under. Following the…

Abstract: This report synthesises findings from 16 peer-reviewed papers addressing the following research question: How does the integration of attention mechanisms in code generation dependency graph models affect robustness against adversarial attacks compared to traditional graph neural networks, measured by. Real-time…

Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: What is the impact of scaling graph size on the adversarial resilience of GCNs for dependency graph imputation in code generation tasks, measured by F1-score degradation under adversarial attacks. Real-time…

Abstract: This report synthesises findings from 7 peer-reviewed papers addressing the following research question: How do multi-view GNN anomaly detection models with view dropout perform on cross-domain datasets compared to single-view models, measured by generalization accuracy and robustness metrics. Anomaly detection is…

Abstract: This report synthesises findings from 4 peer-reviewed papers addressing the following research question: What is the effect of gradient checkpointing on memory efficiency and training throughput in Graph Neural Networks (GNNs) when applied to node classification tasks across datasets with varying graph. Graph Neural…

Abstract: This report synthesises findings from 4 peer-reviewed papers addressing the following research question: How does the sampling rate of node features impact the convergence speed and final accuracy of Graph Attention Networks (GATs) on large-scale heterogeneous graph datasets. We investigate some of the properties and…

Abstract: This report synthesises findings from 12 peer-reviewed papers addressing the following research question: How does the depth of message-passing layers in Graph Neural Networks correlate with accuracy degradation on semi-supervised node classification benchmarks for Cora and Citeseer. Graph Neural Networks (GNNs) have…

Abstract: This report synthesises findings from 13 peer-reviewed papers addressing the following research question: Does GRPO optimization improve the zero-shot generalization of LLaVA-UHD on out-of-distribution multimodal benchmarks compared to PPO. Large vision-language models have achieved outstanding performance, but their…

Abstract: This report synthesises findings from 4 peer-reviewed papers addressing the following research question: How does multi-view graph aggregation affect the pass@k scores of code generation models on the HumanEval benchmark when program dependency graphs are subjected to adversarial node perturbations. Generating…

Abstract: This report synthesises findings from 13 peer-reviewed papers addressing the following research question: How does the robustness of MTGCL with topological features compare to baseline graph contrastive learning methods when subjected to adversarial attacks (e.g., node/edge perturbations) on Ethereum. Real-time…

Abstract: This report synthesises findings from 15 peer-reviewed papers addressing the following research question: Does cross-domain adaptation of data augmentation strategies from image-based models to multimodal language models enhance performance on tasks like VQA and OCR-VQA, and how does this compare to. The field of…

Abstract: This report synthesises findings from 4 peer-reviewed papers addressing the following research question: To what extent do transformer-based anomaly detection models improve cross-domain generalization compared to graph neural networks when measured by AUC-PR on heterogeneous network traffic benchmarks. Anomaly…

Abstract: This report synthesises findings from 15 peer-reviewed papers addressing the following research question: How does the inference efficiency of deep learning-based network intrusion detection models compare to traditional signature-based systems when evaluated on high-throughput benchmark datasets. Cyber intrusion…

Abstract: This report synthesises findings from 15 peer-reviewed papers addressing the following research question: How does adversarial training impact the inference throughput of graph neural networks on high-frequency spatio-temporal data streams compared to standard training. Missing data imputation (MDI) is a fundamental…

Abstract: This report synthesises findings from 3 peer-reviewed papers addressing the following research question: How does the sample efficiency of graph contrastive learning methods like GRACE and DGI compare to supervised approaches when labeled anomalies are limited to less than 5\% of the dataset. Log analysis is one of…

Abstract: This report synthesises findings from 1 peer-reviewed paper addressing the following research question: How does the inference latency of graph-based anomaly detection models compare to transformer-based sequence models when evaluating long-context reasoning benchmarks on large-scale datasets. An AI-driven anomaly…