Browse Articles

Here the authors evaluateE various configurations of Llama 3.2 models (ranging from 1B to 8B parameters trained across 1 to 5 epochs) to serve as cost-effective math tutoring assistants for K-12 classrooms. Their findings show that while the largest 8B model trained for 5 epochs achieved peak accuracy, smaller fine-tuned models provide a viable, resource-efficient balance for practical educational deployment.

In this study, we investigated how the complexity of a query can influence tool-use hallucinations in large language models that were tasked with setting an appointment. Using explainable AI (XAI) as a tool to test the hypothesis, we analyzed trends in hallucination rates by the type of query and saw results that when complexity and ambiguity of the query increased (such as asking for multiple days or unclear time ranges), the model was more likely to make errors due to misunderstanding time-related data from tool outputs. This shows that making large language models (LLMs) understand what seems logical to a human is still difficult, especially in complex tasks, and that knowing what causes these errors can help build more trustworthy AI systems.

The authors looked at ways that autonomous vehicles could be trained to better perform in adverse weather without the use of actual images taken during adverse weather events. They were able to improve performance, but models using actual images still perform the best.

This manuscript focuses on improving the mechanical strength and radiopacity of SLA 3D-printed stent materials by incorporating nanoparticles for their application in flow-diverting stents (FDSs) used to treat aneurysms. The authors used biocompatible, bioresorbable resin formulations with varying concentrations of silica and zirconium to enhance tensile strength and radiopacity, respectively. The authors employed SEM to confirm the incorporation of nanoparticles, and CT imaging was used to assess the effect of zirconium-infused FDSs.

The authors designed a triboelectric nanogenerator with the goal to use it as an energy source for smart farming.

This study investigates the dynamic reaction between liquid gallium and aluminum, which poses a challenge when using high-performance gallium-rich thermal interface materials for electronic cooling. Through the usage of in-situ microscopy, we show that while aluminum oxide coatings slow gallium-induced damage, nanometer-thick iridium coatings effectively prevent reaction and surface degradation at device-operating temperatures. These findings highlight a promising thermal-cooling architecture for extending the lifespan and reliability of high-power electronic heat dissipation systems.

Machine learning systems are often expected to make fair decisions, yet the most widely used fairness criteria—independence, separation, and sufficiency—cannot generally be satisfied at the same time. In this study, students tested these criteria using a logistic regression model on a real-world student performance dataset and found that each criterion was met only at different prediction thresholds, with no threshold satisfying all three simultaneously. These results illustrate the inherent trade-offs in algorithmic fairness and highlight why achieving perfectly fair machine learning models is often impossible in practice.

The authors investigated the efficacy of functionalized graphene oxide nanoparticles for turning saltwater to freshwater.

The purpose of this study was to determine the necessity of previous non-algorithmic attacks (Adversarial Training) in light of algorithmic defense methods (Gradient Masking and Defensive Distillation) against FGSM attacks. We found a significant increase in image classification accuracy from defense methods with the non-algorithmic defense method compared to ones without. By analyzing the significance with a McNemar test, we determined that the inclusion of non-algorithmic defense methods is still necessary in light of new algorithmic defense methods.

This manuscript tackles a major social issue in the health news sector, with social media being one of the primary sources of information and a prime spot to propagate fake news. The author proposes X-HND , which is a unique architecture that combines emotional and contextual analysis in a Graph Neural Network to accurately detect fake news. This was a multi-step process which involved the creation of a custom health news dataset (HNDataset), and an emotional variant that uses RoBERTa to extract emotion. These dataset were then used to prove the hypothesis that accuracy increases when the custom dataset is used to train the model and that with the integration of emotion capture, the detection accuracy increases further.