Mason Journals (George Mason Univ.)
Not a member yet
3256 research outputs found
Sort by
Teaching and Studying the Vietnam War in Vietnamese Higher Education: Current Status and Insights
Get PDF
Optimization of Mechanical Exfoliation Parameters for 2D MoSe₂
No full text2D materials are widely researched due to their advanced physical, electrical, and optical properties, with promising applications in nanoelectronics. Molybdenum Diselenide (MoSe2), a family of transition metal dichalcogenides (TMDs), is an emerging 2D material candidate composed of layered structures of molybdenum and selenium atoms. Currently available exfoliation methods to obtain 2D materials are commonly split between top-down and bottom-up methods: top-down methods focus on exfoliating 2D materials from bulk crystals while bottom-up strategies, such as Chemical Vapor Deposition (CVD), utilize atomic-level interactions to form 2D materials. This study focuses on optimizing the mechanical exfoliation parameters for high yield of 2D MoSe2 flakes. Specifically, different types and combinations of tape (3M Scotch Tape, Nitto SWT 10+ Tape, and Revalpha 3195VS (4M) Thermal Release Tape) were tested alongside thermal treatment ranging from 30℃ to 170℃ and tape filtration. Optical microscopy revealed that Revalpha 3195VS (4M) Thermal Release Tape produced the largest flake size with slight residue as compared with the other two. It was also found that using Revalpha 3195VS (4M) as an intermediary step or thermal treatment at high temperatures helps to increase flake sizes among all tapes, but leaves residue stains on flakes. Finally, two filtrations yielded the highest flake sizes.The results demonstrate the improved parameters of mechanical exfoliation for larger and higher yields of 2D flakes, which can be utilized for nanoelectronics
Enhancing the Performance of Small Language Models in Programming Tasks Through Deep Research and Task Decomposition
No full textRecent advances in large language models (LLMs) have demonstrated powerful capabilities for code generation and reasoning, accelerating development on real-world engineering tasks. However, their high resource demands limit real-world deployment and prevent providers from experimenting with strategies that delve more in-depth and require more token usage. Smaller language models (<33B parameters) present a promising alternative but typically underperform on complex programming tasks. This research addresses the performance gap by creating a system that enhances small language model capabilities through task decomposition and deep research. We developed a modular agent framework using Pydantic to orchestrate multiple LLMs within an isolated container, decomposing programming tasks into discrete subtasks: chunking, planning, research, and implementation. Each subtask undergoes iterative refinement until it meets predefined quality criteria before integration into subsequent stages. The approach was evaluated on a subset of Software Engineering (SWE-bench-lite), which consists of 300 real-world, self-contained functional Python bug fixes derived from GitHub repository issues. The issues involve the use of popular Python libraries like Django and scikit-learn and necessitate the skill of working in large codebases. Preliminary evaluations revealed that our full pipeline outperformed the baseline (without decomposition) in successful task completion rates. The first two Python tasks failed under the baseline with different errors, but completed successfully using the full pipeline. These initial findings suggest that systematic task decomposition and deep research offer a viable pathway for improving small LLM effectiveness in complex programming scenarios, potentially enabling more accessible and cost-effective deployment of AI-assisted software development tools. While the initial research prioritized improving task completion rates, it did not explicitly optimize for execution time. Future research will focus on reducing end-to-end execution time through optimization techniques, such as adaptive decomposition and result caching
Tracking Claude Sonnet 4’s Answers Drift when Explaining Beginner SQL Concepts
No full textThe new rise of AI assistants has completely altered the field of education. Part of what makes AI so effective is its ability to monitor itself and adapt; this strength also introduces a flaw. Although large language models (LLMs) have been proven to be a beneficial tool, there is still a lack of research on the implications that AI drift has on student’s learning. AI drift is when LLMs’ behavior or performance changes over time. This project focuses on quantifying LLM’s drift in its responses to beginner SQL queries. Chen et.al 2024’s previous study found that ChatGPT drifted overtime. We extend the methodology of that longitudinal study to a new test case, focusing on beginner SQL queries, first using a screening step to select the LLM. A list of 6 possible LLMs was created with the criteria that they were conversational, able to read text from images, free, and easily accessible: ChatGPT 4o, DeepSeek, Claude Sonnet 4, Gemini, Meta LLama 3.1, and CoPilot . Then, each LLM was asked 6 questions and graded by 4 high schoolers with limited SQL knowledge on its responses based on a rubric with four categories: accuracy, clarity, wordiness, and pedagogical value. Claude Sonnet 4 was found to provide the best scored responses (18.25/20), so it was utilized to conduct the study. Claude was presented with the same 10 questions every 3 days for 2 weeks. Although limited in test duration, there has been a very slight (1.78%) increase in response quality across all categories. However, the AIA is making the same errors each trial, indicating a possible lack of change in solving potential despite improving the understandability of their answers. Future work would involve a longer time frame and more questions to further confirm the preliminary results.
Chen, L., Zaharia, M., & Zou, J. (2024). How Is ChatGPT’s Behavior Changing Over Time?. Harvard DataScience Review, 6(2). https://doi.org/10.1162/99608f92.5317da4
Large Language Model Selection for Dynamic LLM-GNN Integration
No full textGraph Neural Networks (GNNs) have achieved state-of-the-art results across various domains; however, designing optimal GNN architectures with modifications remains computationally expensive, requiring manual tuning or tedious optimizations. The Prompt-Responsive GNN (PR-GNN) model introduces a novel architecture that leverages real-time LLM feedback to iteratively adjust GNN components based on evolving data or task demands. Given the LLM’s role in this model, selecting the optimal one is essential for maximizing performance. This research proposes an initial framework for comparing four LLMs (ChatGPT-4o, Claude Sonnet 4, Gemini 2.5 Pro, and Grok 3), selected based on current benchmarks. The preliminary methodology included 36 prompts of varying complexities (beginner, intermediate, advanced). These prompts were used to evaluate the LLM’s parsing success. All tested LLMs achieved perfect parsing; therefore, the study adopted an alternative approach that utilized public Kaggle datasets with standardized scikit-learn accuracy metrics. Each LLM generated prompts from the dataset descriptions and code for evaluation in multiple real-world domains, including finance, cybersecurity, and healthcare, to minimize domain-specific bias. Initial results revealed that Claude Sonnet 4 (89.15%) and Grok 3 (89.07%) were statistically equivalent top performers, while Gemini 2.5 Pro and ChatGPT-4o trailed, with 82.86% and 48.63%, respectively. A subsequent assessment of 22 new tasks between Grok 3 and Claude Sonnet 4 revealed nuanced differences (65.73% vs. 62.17%), with Grok 3 slightly outperforming. Additional statistical analysis using 10% trimmed means (86.58% vs 81.55%) confirmed Grok 3's slight advantage by excluding extreme outliers. These findings identify Grok 3 as the preferred choice for integration during further development of the PR-GNN model. The next phase will focus on building a scalable pipeline, assessing performance using fine-grained metrics, and validating Grok 3’s effectiveness through comparative analysis and robustness testing
AKT/mTOR Pathway Protein Expression in the Tumor Microenvironment of High-Risk, HR+, and HER2- Breast Cancer using Laser-Capture Microdissection (LCM) and Reverse Phase Protein Arrays (RPPA)
No full textOver-activation of the AKT/mTOR signaling pathway is one of the main contributors of tumor proliferation and endocrine therapy resistance in breast cancer patients. The tumor microenvironment participates in tumor maintenance and metastasis and may be a key to identify targetable biomarkers and developing personalized treatments. However, current treatments focus mainly on targeting cancer cells and there is still much to learn about how the tumor microenvironment contributes to tumor progression. In this study, pure populations of stroma cells were isolated from twelve HR+/HER2- breast tumor specimens using laser capture microdissection (LCM) and the activation of AKT/mTOR signaling pathway related proteins was quantified and compared to matched whole stroma/tissue lysates utilizing reverse-phase protein arrays (RPPA). Our findings revealed significant differences in 4E-BP1 (S65) (p=0.0003), HER2 (p=0.00001), and elF4E (S209) (p=0.003) in the isolated stroma population compared to the whole tissue lysates where intensity values were higher by 61.6% in 4E-BP1 (S65), 73.7% in HER2, and 86.5% in elF4E (S209), indicating whole tissue lysates were not suitable for a true representation of protein activation. Interestingly, elF4E (S209) was moderately expressed in three of the twelve microdissected biopsies, while all whole tissue lysates showed increased activation. In addition, expression of AKT (S473), AKT (T308) and p70 S6 Kinase (T389) was not detected in isolated stroma or whole tissue lysates. Such differences in protein expression support the rationale that utilizing pure cell populations produces more sensitive results representing the true activation of proteins, which can guide our understanding on how the tumor microenvironment can support development of targeted therapies for high-risk breast cancer patients.