Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: Does cross-domain adversarial training across graph datasets (e.g., citation networks to social networks) improve the robustness of message-passing GNNs and diffusion models, as measured by. 17 claims were…

Abstract: This report synthesises findings from 12 peer-reviewed papers addressing the following research question: How do adversarial robustness improvements in graph diffusion models compare to enhanced message-passing GNNs when measured with robustness metrics on standard graph benchmarks like OGBN-Proteins. 15 claims were…

Abstract: This report synthesises findings from 12 peer-reviewed papers addressing the following research question: How do graph diffusion models and sparse GNN architectures differ in robustness scores against adversarial structural perturbations on large-scale graph benchmarks. 15 claims were extracted from source…

Abstract: This report synthesises findings from 12 peer-reviewed papers addressing the following research question: How do memory consumption and batch processing efficiency differ between graph diffusion samplers and standard GNN encoders when evaluating robustness on graphs exceeding 100k nodes. 0 claims were extracted from…

Abstract: This report synthesises findings from 12 peer-reviewed papers addressing the following research question: How does the inference memory complexity of graph diffusion models compare to sparse GNNs when processing large graphs with high-frequency spectral perturbations. 8 claims were extracted from source literature; 8…

Abstract: This report synthesises findings from 4 peer-reviewed papers addressing the following research question: How does sparse mixture-of-experts routing in multimodal models influence alignment scores on RLHF benchmarks compared to dense baseline architectures. 19 claims were extracted from source literature; 0 were…

Abstract: This report synthesises findings from 4 peer-reviewed papers addressing the following research question: How does adaptive graph pooling affect the robustness of code generation models against adversarial perturbations in the program property graph, as measured by accuracy on the MBPP benchmark. 10 claims were…

Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: What is the impact of dynamic graph pooling on the inference latency of large-scale code generation models when evaluated on the HumanEval benchmark with varying input sequence lengths. 9 claims were extracted…

Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: How do GCN-enhanced models scale in terms of reasoning accuracy on MATH and GSM8K benchmarks compared to non-GCN baselines when evaluated with exact match scores under different model sizes. 0 claims were…

Abstract: This report synthesises findings from 12 peer-reviewed papers addressing the following research question: What is the impact of data augmentation techniques on the accuracy and inference efficiency of multimodal models on VQA and COCO-Caption benchmarks when evaluated with FID and BLEU scores. 18 claims were…

Abstract: This report synthesises findings from 15 peer-reviewed papers addressing the following research question: How do multimodal graph attention networks like MGAT or DGAT compare to traditional GNNs (GCN, GraphSAGE) in terms of inference efficiency on large-scale heterogeneous graphs like PDNS-Net, measured. 17 claims…

Abstract: This report synthesises findings from 16 peer-reviewed papers addressing the following research question: How does the adversarial robustness of GCN-enhanced code generation models compare to transformer-based baselines on HumanEval and MBPP benchmarks when measured by pass@1 under PGD attacks. 0 claims were…

Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: How does the integration of Graph Convolutional Networks (GCNs) in multimodal models affect their robustness against iterative diffusion attacks in language and vision tasks, as measured by accuracy. 10 claims…

Abstract: This report synthesises findings from 15 peer-reviewed papers addressing the following research question: How do different generative AI models (e.g., DALL-E 2, Stable Diffusion) perform in terms of multimodal reasoning accuracy when evaluated on benchmarks like MiniGPT-4 and LLaVA, and how does. 16 claims were…

Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: What is the impact of model size and architecture complexity on the inference efficiency of GCN-enhanced multimodal models when evaluated on downstream tasks such as Hateful Memes and COCO Captions. 0 claims were…

Abstract: This report synthesises findings from 10 peer-reviewed papers addressing the following research question: What is the impact of adversarial graph perturbations on the detection accuracy of contrastive graph neural networks versus autoencoder models in intrusion detection tasks, as evaluated on standard. 0 claims were…

Abstract: This report synthesises findings from 4 peer-reviewed papers addressing the following research question: How do multimodal pose estimation models integrating IMU and camera data compare to Deep Inertial Poser in terms of reconstruction accuracy (MSE) and robustness to occlusions on the 3DPW outdoor. 4 claims were…

Abstract: This report synthesises findings from 11 peer-reviewed papers addressing the following research question: What is the impact of different attention mechanisms (e.g., self-attention, cross-attention) in multi-view GAT models on anomaly detection accuracy and F1 scores in graph-based benchmark datasets. 0 claims were…

Abstract: This report synthesises findings from 16 peer-reviewed papers addressing the following research question: What is the impact of incorporating synthetic IMU data generated by diffusion models on the generalization capability of Deep Inertial Poser, as measured by MSE on unseen human motions in the H36M. 12 claims were…

Abstract: This report synthesises findings from 4 peer-reviewed papers addressing the following research question: How robust are HGT and DAGNN to noise in edge-type annotations when benchmarked on OGB-LSC datasets with varying levels of synthetic edge-type corruption. 12 claims were extracted from source literature; 0 were…

Abstract: This report synthesises findings from 4 peer-reviewed papers addressing the following research question: How does the inference efficiency of multi-view GNNs scale with increasing graph size compared to single-view models when evaluated on benchmark datasets like ACM or DBLP, measured by latency per. 16 claims were…

Abstract: This report synthesises findings from 4 peer-reviewed papers addressing the following research question: How does the computational efficiency of MECCH compare to GAT or GIN in terms of training throughput (edges processed per second) on large-scale heterogeneous graph benchmarks like OGB-LSC or MAG240M. 10 claims…

Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: To what extent do meta-path context convolutions improve robustness against label noise in few-shot learning scenarios compared to standard heterogeneous graph attention networks on benchmark datasets. 0 claims…

Abstract: This report synthesises findings from 11 peer-reviewed papers addressing the following research question: How does pre-training graph neural networks on synthetic fingerprint proxies compare to contrastive learning alignment for improving out-of-distribution generalization in low-data regimes. 12 claims were…

Abstract: This report synthesises findings from 4 peer-reviewed papers addressing the following research question: What is the impact of varying retrieval augmentation strategies (e.g., dense vs. sparse retrieval) on the accuracy and throughput of code generation in the HumanEval benchmark when using LLMs. 7 claims were…