Abstract: This report synthesises findings from 1 peer-reviewed paper addressing the following research question: What is the impact of contrastive meta-learning on the robustness of large language models for few-shot anomaly detection in code when training data contains synthetic noise. 9 claims were extracted from source…

Abstract: This report synthesises findings from 3 peer-reviewed papers addressing the following research question: What is the trade-off between multi-scale feature extraction efficiency and model convergence speed in deep 3D convolutional networks for volumetric data analysis. 9 claims were extracted from source literature; 9…

Abstract: This report synthesises findings from 4 peer-reviewed papers addressing the following research question: How does the inclusion of multimodal context (e.g., code structure visualizations or natural language vulnerability descriptions) in prompting affect Codestral's performance in vulnerability. 17 claims were…

Abstract: This report synthesises findings from 12 peer-reviewed papers addressing the following research question: What is the impact of scaling model parameters on the zero-shot speaker adaptation accuracy of flow-matching speech generators versus autoregressive models. 5 claims were extracted from source literature; 5 were…

Abstract: This report synthesises findings from 5 peer-reviewed papers addressing the following research question: Does the integration of negative sample scaling with low-rank alignment improve robustness against adversarial code-switching attacks in multilingual code generation tasks compared to standard. 11 claims were…

Abstract: This report synthesises findings from 15 peer-reviewed papers addressing the following research question: To what extent does removing language identity via low-rank subspaces degrade zero-shot reasoning capabilities in low-resource languages on multilingual mathematical reasoning benchmarks. 9 claims were extracted…

Abstract: This report synthesises findings from 8 peer-reviewed papers addressing the following research question: How do multimodal models (e.g., combining code and behavioral features) with adversarial fine-tuning compare to unimodal models in terms of robustness against evasion attacks on malware detection. 19 claims were…

Abstract: This report synthesises findings from 12 peer-reviewed papers addressing the following research question: What is the effect of model complexity (e.g., depth, width) on the robustness of adversarial fine-tuned ensemble defenses in malware detection models against evasion attacks, as measured by accuracy. 11 claims…

Abstract: This report synthesises findings from 5 peer-reviewed papers addressing the following research question: How does low-rank subspace projection affect the inference latency and memory footprint of multilingual transformers during cross-lingual retrieval on XQuAD compared to full-rank adversarial. 10 claims were…

Abstract: This report synthesises findings from 10 peer-reviewed papers addressing the following research question: How does the performance of multilingual rumor detection models trained with adversarial contrastive learning compare when fine-tuned on low-resource languages versus high-resource languages, as. 9 claims were…

Abstract: This report synthesises findings from 13 peer-reviewed papers addressing the following research question: How does adversarial fine-tuning with ensemble defenses impact the inference efficiency (e.g., latency, throughput) of malware detection models on large-scale synthetic datasets compared to vanilla. 8 claims were…

Abstract: This report synthesises findings from 5 peer-reviewed papers addressing the following research question: How do adversarial contrastive pre-trained models perform on cross-domain rumor detection tasks, as measured by accuracy on datasets like PHEME and FakeNewsNet, compared to non-adversarial. 12 claims were…

Abstract: This report synthesises findings from 12 peer-reviewed papers addressing the following research question: How do transformer-based multimodal models compare to decoder-only LLMs in cross-modal retrieval tasks on MSCOCO and Flickr30K when using manifold-aware distance metrics versus cosine similarity for. 9 claims…

Abstract: This report synthesises findings from 5 peer-reviewed papers addressing the following research question: How does adversarial contrastive pre-training compare to non-adversarial contrastive pre-training in terms of accuracy and F1-score on the multilingual rumor detection benchmark when evaluated on the. 6 claims…

Abstract: This report synthesises findings from 5 peer-reviewed papers addressing the following research question: What is the impact of model quantization on the inference throughput of adversarial versus non-adversarial contrastive pre-trained multilingual rumor detection models when deployed on edge devices. 7 claims were…

Abstract: This report synthesises findings from 12 peer-reviewed papers addressing the following research question: What is the impact of embedding magnitude normalization (QNorm vs. DNorm) on retrieval accuracy and inference efficiency in multi-hop QA tasks compared to cosine similarity and dot product. 6 claims were…

Abstract: This report synthesises findings from 6 peer-reviewed papers addressing the following research question: How robust are QNorm and DNorm distance metrics to domain shifts in retrieval tasks when evaluated on a diverse set of LLM-based encoders (e.g., BERT, RoBERTa, T5) across BEIR benchmarks. 8 claims were extracted…

Abstract: This report synthesises findings from 13 peer-reviewed papers addressing the following research question: How does the dynamic attention mechanism in multimodal legal question-answering systems influence robustness against adversarial attacks compared to static attention baselines, as measured by. 9 claims were…

Abstract: This report synthesises findings from 10 peer-reviewed papers addressing the following research question: Can cross-encoder models with manifold-aware objectives maintain competitive accuracy on standard retrieval tasks (e.g., SQuAD, TriviaQA) while improving performance on adversarial benchmarks, as. 10 claims were…

Abstract: This report synthesises findings from 15 peer-reviewed papers addressing the following research question: How does the integration of private contextual data affect the retrieval performance of dense models on public benchmarks like BEIR, when evaluated using recall@100 and MRR, and what trade-offs exist. 10 claims…

Abstract: This report synthesises findings from 15 peer-reviewed papers addressing the following research question: Do manifold-aware cross-encoder models achieve higher robustness on adversarial retrieval benchmarks like Adversarial NQ compared to traditional dense retrievers, as measured by MRR@10 and recall. 13 claims were…

Abstract: This report synthesises findings from 7 peer-reviewed papers addressing the following research question: What is the memory footprint comparison between manifold-aware and Euclidean-based models when deployed on edge devices for real-time object detection tasks using benchmarks like COCO-2017. 6 claims were extracted…

Abstract: This report synthesises findings from 10 peer-reviewed papers addressing the following research question: How do manifold-aware fine-tuning techniques impact the inference throughput (queries per second) compared to Euclidean-based models on adversarial image detection benchmarks such as ImageNet-Adv. 12 claims were…

Abstract: This report synthesises findings from 15 peer-reviewed papers addressing the following research question: What are the performance gains of fine-tuning dual-encoder retrievers with domain-specific manifold-aware loss functions on cross-domain natural language understanding benchmarks such as GLUE or. 13 claims were…

Abstract: This report synthesises findings from 12 peer-reviewed papers addressing the following research question: What is the computational efficiency trade-off between momentum adversarial domain-invariant representations and traditional domain-adaptive models during inference in cross-domain retrieval tasks. 11 claims were…