Searcharxiv⌕ Search

arXiv subjects

Kira Sam

Publications and source records attributed to Kira Sam.

5 recordsLinked to original sources
arXiv2024

TBBC: Predict True Bacteraemia in Blood Cultures via Deep Learning

Bacteraemia, a bloodstream infection with high morbidity and mortality rates, poses significant diagnostic challenges. Accurate diagnosis through blood cultures is resource-intensive. Developing a machine learning model to predict blood culture outcomes in emergency departments offers potential for improved diagnosis, reduced healthcare costs, and mitigated antibiotic use.This thesis aims to identify optimal machine learning techniques for predicting bacteraemia and develop a predictive model using data from St. Antonius Hospital's emergency department. Based on current literature, CatBoost and Random Forest were selected as the most promising techniques. Model optimization using Optuna prioritized sensitivity.The final Random Forest model achieved an ROC AUC of 0.78 and demonstrated 0.92 sensitivity on the test set. Notably, it accurately identified 36.02% of patients at low risk of bacteraemia, with only 0.85% false negatives.Implementation of this model in St. Antonius Hospital's emergency department could reduce blood cultures, healthcare costs, and antibiotic treatments. Future studies should focus on external validation, exploring advanced techniques, and addressing potential confounders to ensure model generalizability.

cs.LG↗
arXiv2024

A Comparative Analysis on Ethical Benchmarking in Large Language Models

This work contributes to the field of Machine Ethics (ME) benchmarking, which develops tests to assess whether intelligent systems accurately represent human values and act accordingly. We identify three major issues with current ME benchmarks: limited ecological validity due to unrealistic ethical dilemmas, unstructured question generation without clear inclusion/exclusion criteria, and a lack of scalability due to reliance on human annotations. Moreover, benchmarks often fail to include sufficient syntactic variations, reducing the robustness of findings. To address these gaps, we introduce two new ME benchmarks: the Triage Benchmark and the Medical Law (MedLaw) Benchmark, both featuring real-world ethical dilemmas from the medical domain. The MedLaw Benchmark, fully AI-generated, offers a scalable alternative. We also introduce context perturbations in our benchmarks to assess models' worst-case performance. Our findings reveal that ethics prompting does not always improve decision-making. Furthermore, context perturbations not only significantly reduce model performance but can also reverse error patterns and shift relative performance rankings. Lastly, our comparison of worst-case performance suggests that general model capability does not always predict strong ethical decision-making. We argue that ME benchmarks must approximate real-world scenarios and worst-case performance to ensure robust evaluation.

cs.CY↗
arXiv2024

CELLM: An Efficient Communication in Large Language Models Training for Federated Learning

Federated Learning (FL) is a recent model training paradigm in which client devices collaboratively train a model without ever aggregating their data. Crucially, this scheme offers users potential privacy and security benefits by only ever communicating updates to the model weights to a central server as opposed to traditional machine learning (ML) training which directly communicates and aggregates data. However, FL training suffers from statistical heterogeneity as clients may have differing local data distributions. Large language models (LLMs) offer a potential solution to this issue of heterogeneity given that they have consistently been shown to be able to learn on vast amounts of noisy data. While LLMs are a promising development for resolving the consistent issue of non-I.I.D. Clients in federated settings exacerbate two other bottlenecks in FL: limited local computing and expensive communication. This thesis aims to develop efficient training methods for LLMs in FL. To this end, we employ two critical techniques in enabling efficient training. First, we use low-rank adaptation (LoRA) to reduce the computational load of local model training. Second, we communicate sparse updates throughout training to significantly cut down on communication costs. Taken together, our method reduces communication costs by up to 10x over vanilla LoRA and up to 5x over more complex sparse LoRA baselines while achieving greater utility. We emphasize the importance of carefully applying sparsity and picking effective rank and sparsity configurations for federated LLM training.

cs.LG↗
arXiv2024

UAV Networks Surveillance Implementing an Effective Load-Aware Multipath Routing Protocol (ELAMRP)

In this work uses innovative multi-channel load-sensing techniques to deploy unmanned aerial vehicles (UAVs) for surveillance. The research aims to improve the quality of data transmission methods and improve the efficiency and reliability of surveillance systems by exploiting the mobility and adaptability of UAVs does the proposed protocol intelligently distribute network traffic across multiple channels, considering the load of each channel, While addressing challenges such as load balancing, this study investigates the effectiveness of the protocol by simulations or practical tests on The expected results have improved UAV-based surveillance systems, more flexible and efficient networks for applications such as security, emergency response and the environment alignment of monitoring -Offering infrastructures, which contribute to efficient and reliable monitoring solutions.

cs.NI↗