Computational Research & Cyberinfrastructure Working Group
đź“Ť Library North, Classroom 1
🗓 Last Friday of each month | 🕦 11:30 AM
Please note that this month event will be held on Thursday, May 29th
We invite you to present at the Computational Research & Cyberinfrastructure Working Group! This is a great opportunity to:
- Share your knowledge with peers and gain presentation experience
- Build your academic and professional profile by showcasing your expertise
- Expand your network while enjoying a free lunch!
- Contribute to the research community by helping others improve their computational skills
Simplified and Branching of Deposition and Erosion in Porous Media
Amy MarĂa Sims, Ph.D. student in Mathematics and Statistics
(Advisor: Dr. Pejman Sanaei)
Erosion and deposition are prevalent and integral phenomena in natural and industrial settings such as agriculture, dam construction, and fluid filtration, among others. Deposition affects the integrity of filters in industrial settings, and erosive activity can deplete soil of nutrients, thus affecting the growth of nearby vegetation. As such, we aim to develop mathematical models that describe and predict the behavior of porous media subject to these naturally occurring processes. We examine values and limitations of parameters such as porosity, shear stress, particle concentration, and porous medium deformation. By exploiting established governing equations such as the Darcy, advection-diffusion-reaction, and Navier-Cauchy equations, and then simplifying calculations via nondimensionalization and asymptotic analysis based on the small aspect ratio of the medium, we successfully create models from both macroscale and microscale perspectives that depict how the erosion and deposition processes alter the internal morphology of an elastic porous medium. The macroscale, simplified continuum model assumes homogeneity at each cross-section of the medium, and the microscale, branching model assumes homogeneity of each branching structure along the length and depth of the medium, thereby allowing for reduction from three to two (for the continuum model) and one (for the branching model) dimensions. The essence of our results, beyond displaying how the medium evolves subject to the complex interplay among multiple parameters, covers how the total volume of the medium changes as we vary quantities of coefficients that measure the tendency of particles to erode or become deposited based on the physical properties of those particles and of the fluid and medium in which they travel. Further, we analyze how the total volume of the medium changes as it expands and contracts due to deformation induced by increasing or decreasing pressure, respectively.
GraphQL Implementation in the DMAAG Project: A Modern Approach to Historical Data Access
Shreya Tadipaneni, M.S. student in Computer Science
(Advisors: Dr. Elizabeth West & Dr. Chetan Tiwari)
This presentation explores the implementation of GraphQL in the Data Mining and Mapping Antebellum Georgia (DMMAG) project. It highlights the development of a full-stack application that integrates PostgreSQL for database management, GraphQL for API development, and React for the frontend interface. The talk explains how this technology stack enables efficient querying of complex historical records about enslaved individuals in Georgia, outlines the advantages of GraphQL over traditional REST APIs, and demonstrates how the React frontend provides an intuitive, responsive user experience. The presentation showcases practical applications of modern web technologies that can support researchers working with complex datasets across various disciplines.
Toward Agentic AI at Georgia State: A Case Study in Generative Personalization for Student Success
Jaroslav Klc, Director, Strategic Initiative, Instructional Innovation and Technology (IIT)
This case study presentation showcases Georgia State University’s (GSU) innovative proof-of-concept use of generative AI to transform how students engage with financial aid and academic support services. In response to the complexity and confusion that often surround financial aid processes, GSU developed a personalized, AI-powered dashboard that functions as a virtual financial aid coordinator. Built using Amazon Bedrock with the Anthropic Claude V2 foundation model, ElasticSearch vector databases, and Elastic’s semantic search engine (ELSER), the system delivers context-sensitive, student-specific content derived from institutional data systems.
By dynamically prioritizing tasks such as deadlines, eligibility criteria, and registration requirements, the platform ensures that students receive only the information that applies to their unique academic and financial profile—improving clarity, engagement, and completion rates. The generative AI approach enables scalable delivery of tailored guidance, augmenting the capacity of human advisors while streamlining communication.
Looking ahead, GSU IIT plans to expand this architecture using agentic AI, introducing autonomous agents that emulate key student support roles—including academic advisor, career coach, health advisor, and financial coach. These agents will collaborate behind the scenes to deliver holistic, hyper-personalized context-specific content and recommendations, offering students an integrated support experience that adapts in real time to their evolving needs and goals.
Leveraging PySpark in Large Scale Solar Transient Event Prediction Dataset Construction
Dustin J Kempton, Assistant Professor, Computer Science
NASA’s Solar Dynamics Observatory (SDO) has produced petabytes of data over it’s operational lifetime, and creating benchmark machine learning datasets for solar transient event prediction from these observations of the Sun is a non-trivial problem. Luckily, when performing data science tasks on large datasets, it is often the case that computations can be broken down into what is commonly referred to as an “embarrassingly parallel workload.” These problems have minimal or no dependency upon communication between the parallel tasks, and can easily be distributed across compute nodes in server farms, or internet-based volunteer computing platforms such as BOINC. In this talk, we will discuss using PySpark to distribute feature engineering calculations on hyper-spectral remote sensing images coming from SDO’s Atmospheric Imaging Assembly. By utilizing this large-scale data processing engine, we can abstract away issues such as determining what compute node is ready for more work units, when a computation needs to be rescheduled to another compute node, dynamically scaling the number of compute workers as processing needs change, etc. Though Spark and PySpark are quite mature, and can automatically optimize many tasks on tabular data, the user still needs to be cognizant of its limitations. For example, Sparks tendency to perform all operations on in memory data. Therefore, we will discuss using the Spark engine to distribute metadata for retrieving the actual data that will have computations performed on them. In doing so, we will also discuss some drawbacks of this, and how optimizing how data is distributed can keep from overwhelming your system, improving your algorithms performance.
Upcoming Events
We offer workshops regularly. See the list below.