Abstract: This report synthesises findings from 16 peer-reviewed papers addressing the following research question: What trade-offs exist between task complexity (e.g., simple vs. high-dimensional control) and the computational efficiency of neuroevolution algorithms when optimizing for both QD-score and maximum. Soft robotics…

Abstract: This report synthesises findings from 13 peer-reviewed papers addressing the following research question: To what extent does EVOR's diverse knowledge base adaptation improve pass@k scores on out-of-domain programming tasks relative to single-source retrieval methods. Recently the retrieval-augmented generation (RAG)…

Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: How do different behavioral descriptor designs in multimodal QD benchmarks influence the robustness of neuroevolution-trained agents when transferred to unseen environments, as evaluated by coverage. We present a…

Abstract: This report synthesises findings from 15 peer-reviewed papers addressing the following research question: How does the accuracy-latency trade-off of distilled FLAN-T5 models vary across student-teacher size ratios when evaluated on the SNLI and MultiNLI benchmarks. Large Language Models achieve remarkable performance…

Abstract: This report synthesises findings from 12 peer-reviewed papers addressing the following research question: How does the integration of multimodal Quality-Diversity (QD) benchmarks (e.g., vision + control) impact the generalization performance of neuroevolution-trained agents in out-of-distribution. We present a…

Abstract: This report synthesises findings from 16 peer-reviewed papers addressing the following research question: How does the robustness of policies evolved via quality-diversity algorithms compare to standard PPO when evaluated under adversarial perturbations in continuous control tasks. The increasing importance of robots…

Abstract: This report synthesises findings from 16 peer-reviewed papers addressing the following research question: How do quality-diversity neuroevolution algorithms compare to gradient-based RL methods in terms of sample efficiency and final reward on MuJoCo locomotion benchmarks. Achieving fast and stable off-policy…

Abstract: This report synthesises findings from 8 peer-reviewed papers addressing the following research question: What is the impact of LongLoRA's shifted sparse attention mechanism on pass@1 scores in HumanEval compared to full fine-tuning and adapter-based methods. We present LongLoRA, an efficient fine-tuning approach that…

Abstract: This report synthesises findings from 11 peer-reviewed papers addressing the following research question: How does the performance degradation of Codestral and DeepSeek R1 on LiveCodeBench compare when evaluated on time-contaminated versus contamination-free coding problems. Large Language Models (LLMs) applied to…

Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: How does the integration of DeepSeek R1 in Scalene affect the robustness of generated optimization suggestions across different Python code domains (e.g., numerical computing, web frameworks), as. Python's…

Abstract: This report synthesises findings from 7 peer-reviewed papers addressing the following research question: What is the impact of quantization techniques (e.g., 4-bit, 8-bit) on the inference efficiency (throughput, latency) of DeepSeek R1 when generating optimized Python code suggestions compared to. The growing demand…

Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: What is the impact of varying the number of retrieved passages on RETRO's performance in code generation tasks measured by HumanEval execution accuracy. Large language models have shown remarkable aptitude in…

Abstract: This report synthesises findings from 4 peer-reviewed papers addressing the following research question: What is the impact of Vendi-RAG's relevance-diversity tradeoff on the adversarial robustness of FLAN-T5-xl against non-lexical overfitting in NLI tasks. Recent advances in large language models (LLMs) have opened…

Abstract: This report synthesises findings from 15 peer-reviewed papers addressing the following research question: How does retrieval diversity in RAG-augmented FLAN-T5-xl affect accuracy on syntactic heuristics in the HANS benchmark compared to standard retrieval. Retrieval-augmented generation (RAG) enhances large language…

Abstract: This report synthesises findings from 11 peer-reviewed papers addressing the following research question: How does the token generation throughput of U-PaLM variants compare to dedicated vision-language models like Flamingo on the VQA v2 dataset across different parameter scales. Vision-Language Model (VLM) have…

Abstract: This report synthesises findings from 4 peer-reviewed papers addressing the following research question: How does the memory efficiency of Longformer-En scale with context length in multimodal document understanding tasks relative to its performance on text-only long-context benchmarks. Reasoning over long sequences…

Abstract: This report synthesises findings from 15 peer-reviewed papers addressing the following research question: Does context-aware chunking improve answer exact match scores on multi-hop QA datasets compared to fixed-size segmentation for transformer models with extended attention spans. Multi-hop question answering is a…

Abstract: This report synthesises findings from 12 peer-reviewed papers addressing the following research question: How do memory consumption patterns differ between ETC, Longformer, and BigBird during training on extended sequence lengths in the HotpotQA benchmark. Transformers-based models, such as BERT, have dramatically…

Abstract: This report synthesises findings from 10 peer-reviewed papers addressing the following research question: How does the inference throughput of ETC compare to Longformer and BigBird on the HotpotQA dataset when sequence lengths exceed 8,000 tokens. Transformers-based models, such as BERT, have dramatically improved…

Abstract: This report synthesises findings from 10 peer-reviewed papers addressing the following research question: How does hierarchical chunking compare to sliding window strategies in preserving long-range reasoning accuracy for Longformer on the HotpotQA benchmark. The effectiveness of Retrieval-Augmented Generation (RAG)…

Abstract: This report synthesises findings from 16 peer-reviewed papers addressing the following research question: What is the impact of structured input encoding in ETC versus sparse attention mechanisms in Reformer on multi-hop reasoning accuracy for long-context question answering. Transformer models have advanced the…

Abstract: This report synthesises findings from 15 peer-reviewed papers addressing the following research question: How does MA-DPR affect retrieval latency and GPU memory consumption compared to cosine similarity in low-resource language settings on XQuAD. Dense Passage Retrieval (DPR) typically relies on Euclidean or cosine…

Abstract: This report synthesises findings from 13 peer-reviewed papers addressing the following research question: What is the impact of MA-DPR on the robustness of dense retrieval against adversarial query perturbations in multilingual QA benchmarks. Dense retrieval has become the new paradigm in passage retrieval. Despite…

Abstract: This report synthesises findings from 13 peer-reviewed papers addressing the following research question: Does replacing cosine similarity with MA-DPR improve cross-lingual zero-shot transfer performance on TyDi QA for languages unseen during training. Large Language Models (LLMs) have demonstrated remarkable…

Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: How does the lexical injection approach proposed in this paper affect the inference efficiency of cross-lingual retrieval models when deployed on edge devices, as measured by latency and throughput. Effective…