Poster Presentations

Using latent profile analysis and a random-intercept cross-lagged panel analysis to examine childhood adversity, reward processing, and psychopathology. 

Last semester I worked on my senior capstone project with the National Center of Women's Innovations where my part was to build a chatbot that could go into their database and answer questions on women innovators. I would like to present about the way I accomplished this.

This project develops automated methods to detect anomalous flight behaviors in ADS-B data. We focus on identifying unusual holding and loitering patterns (Type A anomalies) that deviate from standard air traffic procedures. Using a combination of rule based filtering and machine learning techniques, we analyze spatial, temporal, and kinematic features to distinguish between normal operations and suspicious flight patterns. Our hybrid detection system successfully identifies aircraft exhibiting extended circular patterns, unusual altitude distance relationships, and prolonged presence in confined areas. While the approach effectively flags kinematic anomalies, it cannot determine intent or authorization status, highlighting the need for integrating additional contextual data sources for complete threat assessment.

We present a hybrid conversational chatbot that enables non-expert users to query the U.S. Fatality Analysis Reporting System (FARS) using natural language. While FARS contains decades of detailed fatal- crash data, its complex relational schema and technical SQL requirements limit accessibility for policymakers and researchers. Our system provides an intuitive interface that converts user questions into data-grounded, verifiable responses, improving accessibility, transparency, and efficiency in transportation safety analysis.

Understanding transient ground deformation driven by magmatic activity requires separating short-lived volcanic signals from long-term tectonic and other non-volcanic processes. At active volcanoes such as Ol Doinyo Lengai, located in the Natron Rift segment of the East African Rift System, this separation is critical for interpreting magma transport and assessing volcanic hazards. We integrate 2.5 years of continuous Global Navigation Satellite System (GNSS) data from the TZVOLCANO Global Navigation Satellite System (GNSS) network with 2.5 years of Sentinel‑1 Interferometric Synthetic Aperture Radar (InSAR) data to characterize the spatial and temporal evolution of transient deformation. Sentinel‑1 data (ascending) are obtained through Hybrid Pluggable Processing Pipeline (HyP3) from Alaska Satellite Facility and analyzed using the Miami InSAR Time-series Software in Python (MintPy) to derive Line-of-Sight displacement time series and deformation maps. GNSS data are processed with GAMIT/GLOBK to obtain high-precision station position time series. The Robust Network Imaging algorithm is used to interpolate vertical GNSS-derived velocities using median statistics.  We then jointly combine the GNSS–InSAR observations to create a new, consistent vertical velocity solution of the region and assess deformation.

The Targeted Projection Operator (TPO) is a method that analyzes Global Navigation Satellite System (GNSS) time-series data, which can be used to detect transient deformation signals occurring at a volcano; such transient signals may be a precursor for volcanic activity. Ol Doinyo Lengai, an active volcano in Tanzania located in the East African Rift System, has been continuously monitored by several GNSS stations since 2016. We conduct a time-series analysis through December 2024 to evaluate how surface motions have evolved since a known 2022-2023 uplift episode. Our results suggest a period of subsidence that may be indicative of subsurface depressurization of a magma reservoir beginning in March 2023 and continuing through the end of the data analysis period in December 2024.

Acute Myeloid Leukemia (AML) is one of the most common Leukemias in adults and often needs to be diagnosed and treated right away [1]. Our sponsor, Dr. Ifeyinwa Obiorah has tasked us this semester to aid in the diagnosing process of AML. For a medical professional, this process entails manually counting the proportion of Myeloblasts, a specific type of White Blood Cell, (WBC) out of all WBCs given a bone marrow microscopic image. This motivated the goal of our project: use machine learning (ML) and computer vision techniques to build a model that can classify if a given set of slide images comes from a patient with AML. The data pipeline we developed to achieve this contains two main sub-processes. We first segment a given bone marrow image into sub-images, where each sub-image contains only one cell. Then, we take the set of sub-images, isolate and classify the type of WBC each remaining sub-image is. Last, we calculate the proportion of all WBCs that were classified as Myeloblasts; > 20% is indicative of AML. We tested this count against pseudo test data, as there were not readily available bone marrow images with several WBCs that include the true Myeloblast count. Our testing showed our model yielded an average difference from the true cell count of about ± 2 cells. In all, we believe our project yields promising results for using ML to aid in the diagnosing process of AML while recognizing that our current pipeline is not ready for full deployment in the medical practice setting.

This capstone project, completed in partnership with Kelsey Vinson, Director of Sports Science for Virginia Tech Football, investigated how in-season performance and movement data can inform weekly practice structure and support improved game-day readiness. Using force-plate countermovement jump metrics, eccentric hamstring strength measures, Catapult GPS outputs, game outcomes, and Pro Football Focus (PFF) grades, our team evaluated which routinely collected practice variables most meaningfully relate to competitive performance.

This project focuses on creating a software-defined underwater acoustic modem using GNU Radio with the Raspberry Pi and Red Pitaya platform. The goals include getting DSP code running natively on embedded hardware, developing reliable BFSK modulation/demodulation, and performing controlled BER vs SNR testing in water.​

Machine-generated probability predictions are essential in modern classification tasks such as image classification. A model is well calibrated when its predicted probabilities correspond to observed event frequencies. Despite the need for multicategory recalibration methods, existing  methods are limited to (i) comparing calibration between two or more models rather than directly assessing the calibration of a single model, (ii) requiring under-the-hood model access, e.g., accessing logit-scale predictions within the layers of a neural network, and (iii) providing output which is difficult for human analysts to understand. To overcome (i)-(iii), we propose Multicategory Linear Log Odds (MCLLO) recalibration, which (i) includes a likelihood ratio hypothesis test to assess calibration, (ii) does not require under-the-hood access to models and is thus applicable on a wide range of classification problems, and (iii) can be easily interpreted. We demonstrate the effectiveness of the MCLLO method through simulations and three real-world case studies involving image classification via convolutional neural network, obesity analysis via random forest, and ecology via regression modeling. We compare MCLLO to four comparator recalibration techniques utilizing both our hypothesis test and the existing calibration metric Expected Calibration Error to show that our method works well alone and in concert with other methods.

Vertical land motion (VLM) has been observed for decades in the Chesapeake Bay region, but rates of VLM vary spatially and may be influenced by a variety of local processes. The West Point Land Motion Observatory, maintained by the U.S Geological Survey, is host to a suite of instruments that we use to observe deformation both within and at the surface of the lithosphere providing an ideal place to examine the roles of potential VLM drivers. This ongoing study leverages campaign Global Navigation Satellite System (GNSS) observations, Interferometric Synthetic Aperture Radar (InSAR) observations from Sentinel-1, extensometer data, groundwater level data, and near-surface geophysical data from West Point, Virginia to estimate the local rate of VLM and investigate the roles of potential VLM drivers.

At the core of Projectr is a geospatial intelligence engine built to understand how the physical world actually behaves. The Projectr Engine continuously mines and ingests data from thousands of local, county, state, and federal sources—along with satellite imagery and other spatial signals—and normalizes that information across space and time.

The Chesapeake Bay is experiencing impactful relative sea-level rise driven, in part, by subsiding vertical land motion (VLM) from both natural and human processes. Previous studies have produced conflicting VLM results due to differing methods and datasets. This project represents a detailed analysis of the 2024 Global Navigation Satellite System (GNSS) campaign data, which was collected to evaluate VLM for the region. Data from 51 surveyed benchmarks (48 usable stations) are currently being analyzed.