Mason Journals (George Mason Univ.)
Not a member yet
3256 research outputs found
Sort by
Oddball Questions as a Means to Educate and Energize New World History Teachers: Vietnam, a Case Study
No full text
Assessing the Consistency of Open-Source Large Language Models for Algorithm Evaluation
No full textThe grading of open-ended questions in education is labor-intensive and subject to human error, making it an attractive target for automation through AI. Manual scoring from professionals, although thorough, is a time-consuming task that often lacks consistency and contains bias across algorithms and evaluators. Recent advancements in AI, particularly in large language models (LLMs) like GPT-4, have shown significant promise in this domain. Automated scoring methods require large amounts of training data to ensure generalizability, cost users greatly, and currently are not widely used for grading complex assignments. We tested various LLM prompting strategies such as Chain-of-Thought and comparative grading, and found that rubric-based grading has proven to offer transparency in grading, thorough feedback, and accurate scores compared to human grading. This work explores the consistency of four LLM software – Anthropic’s Claude, OpenAI’s ChatGPT, Microsoft Copilot, and Google Gemini – using rubric grading. Custom rubrics were developed to cover four criteria: algorithm design, completeness, clarity/readability, and logic. We tested the LLMs to provide feedback and grading for one program that intentionally contained various errors. For each LLM, four prompts, one for each rubric category, were inputted 42 times, resulting in a total of 672 data points along with the corresponding feedback. Statistical methods, including analysis of standard deviations and Intraclass Correlation Coefficient (ICC), were employed to evaluate the consistency of LLM grading and feedback per rubric category. The ICC values were utilized to assess the reliability of LLM grading across multiple trials, with high values indicating higher consistency
Computational Identification and Mapping of Protein-DNA Interactions
No full textProtein-DNA interactions play a crucial role in key biological processes such as gene regulation, replication, and transcription. Understanding these interactions strongly influences the development of gene editing technologies like CRISPR. Machine learning models such as AlphaFold and RoseTTAFold can generally predict protein-DNA structures, but models that can explain their critical binding interactions based on protein or DNA sequences alone remain underdeveloped. Enhancing our understanding of these key interactions will improve the accuracy of such models to the point that we can predict ideal protein binding partners for a given DNA sequence. We developed a Python script to explicitly identify key interactions between protein and DNA in 89 protein-DNA interfaces from the Protein Data Bank. Our script facilitates visualization and analysis of interaction patterns along the DNA sequence, offering a robust framework for understanding the key intermolecular interactions underlying known protein-DNA interfaces, and validating the presence of key interactions on DNA bases. From here, we started developing a second algorithm that determines the best-known DNA-binding protein for an arbitrary stretch of DNA. These algorithms could be adapted to generate modified CRISPR machinery to directly recognize specific DNA sequences, thus potentially increasing CRISPR specificity while decreasing dependence on gRNA
Controlled synthesis of gold nano bipyramids
No full textGold nano bipyramids (Au NBPs) and nanorods (Au NRs) have been heavilyresearched due to their unique properties, but the current growth methods have shownto be too tedious. They are specialized gold nanoparticles with distinct shapes, andthere are two common types of elongated colloidal plasmonic metal nanocrystals. Thedifference is that Au NPBs have narrower shapes and sharper tips than Au NRs. Inthis study, we utilizedthe seed-mediated approach to synthesize the Au NBPs sampleand focus on improving the yield of Au NBPs and aim to understand the effect of theconcentration ratio between CTAC/sodium citrate by changing the concentration ofsodium citrate to obtain a high yield of Au NBP. We also tried various CTABconcentrations within our growth solution during our study. We kept most of ourgrowth solution formula the same throughout this experiment but with variedconcentrations of CTAB with the same seed solutions. We had to be precise with howmuch of each chemical we added to our solution and the order in which we addedthem. Preliminary results showed that the growth solution with 40mM of CTABcombined with the concentration CTAC/CiNa3 21:1 seed solution was better than theone with 50mM of CTAB. Each sample's characterization was measured usingTransmission Electron Microscopy (TEM) and UV-Vis spectra. As we find ways toimprove the meditated seed growth of Au NBPs, they have a high potential to beuseful for future uses, such as in the photocatalyst and medical fields
Formal Text Translation Using Large Language Models: A Replication Study on LLM-Driven Translation of Open-Source Software Documentation from English to German Language Variants
No full textOpen-source software development is a global activity, with the potential for engaging a wide array of contributors. Central to contributing in open-source is the documentation made available. With the recent emergence of generative AI and Large Language Models (LLMs), translation across languages using these new tools has been a prominent subject in research. However, studies commonly focus on conversational language, overlooking formal and legal texts which would improve the engagement of global audiences through accurate translation of open-source documentation. Furthermore, most works have focused on translation within global languages rather than across different languages. This study aims to evaluate the performance of LLM translation in formal contexts, specifically open-source documentation. To achieve this goal, we designed a replication study (based on “LLMs and Translation: different approaches to localization between Brazilian Portuguese and European Portuguese”) where we are evaluating different approaches to localization between language variants by using an existing dataset of onboarding documents. However, instead of translating English to Portuguese, we focus on translation from English to German language variants. These efforts have implications for both research and practice, providing insights into the effectiveness of LLMs translating formal, or in our case technical, documents. This could not only increase the accessibility of open-source documentation for global audiences, but also improve the translation of similar, technical texts
An Investigation into the use of Newspaper Data and Large Language Models for Understanding the Occurrence of Flooding in the Caribbean
No full textFlooding is the most common and widespread weather-related natural disaster, causing over $10 billion in property damage and countless losses of lives and livestock globally. Governments and institutions often struggle with inadequate flood protection and relief, particularly in low-and-middle-income countries (LMICs), where nearly 90% of those affected reside. Although early warning systems and infrastructure projects are implemented as solutions, they are often prohibitively expensive for LMICs. A more cost-effective solution is the development of timely flood maps, which relies on up-to-date flood data. Traditional methods, such as deploying large survey teams, are costly and impractical, while satellite imagery and radar data are often hindered by cloud cover and interpretative challenges.
Newspaper reports on flooding, however, present an underutilized data source. They are widely available, have extensive archives, and can be curated online. In this study, we utilize newspaper data from Trinidad and Tobago, a small Caribbean Island where flooding remains a significant issue. Using large language models (LLMs) - ChatGPT, Phi3, and LLaMa - we extract information on flood locations and causes reported in online newspapers. This data is then used to create an updated flood risk map, reflecting the frequency and causes of flooding at the community level. Our results show the most at-risk communities and their flooding causes. The comparison of LLMs demonstrates an accuracy range of 89% to 94% in identifying flood locations, with differences within 10%. This approach enhances flood mapping by providing current, actionable data to governments and institutions, improving flood management and response strategies
Enhancing Human-Robot Collaboration Through Robotic Eye Gaze: A Study on External Stimuli and Predictive Cues in Interaction
No full textIn the era of Industry 4.0, robotics and AI are rapidly evolving to enhance human-robot collaboration. However, factories experience workflow delays and underutilize advanced technologies due to inefficient human-robot integration. Most existing research overlooks human biases and has not explored whether humans will collaborate with robots when guided by external stimuli. This study addresses this gap by investigating whether robotic eye gaze can enhance human-robot collaboration. To explore human behavior with external cues, 51 participants engaged with a six-degrees-of-freedom robot to choose the same buzzer button among two options simultaneously. Participants also performed cognitive and coordination tasks to imitate an industrial environment. Using computer vision, the program analyzed hand angles and applied an opinion dynamics algorithm to establish real-time opinions of both the human and robot. Initially, the robot’s opinion was programmed to actively disagree with the human, stimulating a need for external collaboration. Later, a realistic robotic head used eye movement to indicate its intended actions, with the gaze becoming more pronounced each round. Findings revealed that 91% of participants noticed the robot's eye gaze and used it to influence their decisions, and 86% felt more aligned with the robot's decisions after the eyes began moving. In the final iteration, 100% of participants aligned their button choice with the robot's, demonstrating the effectiveness of visual cues. These results suggest that humans are highly susceptible to influence and bias from a robot to achieve a common goal, highlighting the potential for enhancing human-robot collaboration through subtle external stimuli
Developing an ultra-lightweight capturing solution for organic, compliant bodies.
No full textIn recent years, the field of robotics has seen significant advancements in developing mechanisms for indexing irregular shapes. Despite such advancements, current mechanisms often fail to effectively balance the trade-offs between ultra-lightweight design and the necessary durability and the aerodynamic stability required for airborne applications. Additionally, existing solutions struggle to adapt dynamically to capture and hold irregular shapes efficiently, leading to inflexible performance in real-world scenarios. There is a need for an effective and efficient tool that seamlessly integrates these stringent requirements for dynamic target retrieval. In order to optimize our design for the specific purpose of mounting to DTR blimps, we explored many different low-density materials, geometric outlines, and actuation methods. In order to verify our designs we plan on pairing fluid dynamic simulations with physical scale testing. Our research develops a mechanism that minimizes impact on system weight and aerodynamics while maintaining rigidity and reliability of the design to be used in DTR