33rd USA/USM/SELU Research Mini-Conference

The USA/USM/SELU Mini-Conference on Undergraduate Research in Science and Mathematics is a (very) regional one-day meeting, featuring research presentations in the physical and mathematical sciences from students and faculty from universities in the Gulf Coast area. A principal goal of this conference is to involve undergraduate students in research and provide an opportunity for them to present their work. There is no conference fee, and a free lunch will be provided to all participants.

We encourage interested faculty and students, including those who may not want to present, to attend and learn more about research going on in our Gulf area. For more information, please contact one of the local organizers:

• Arjun Dahal (Physics)
• Martin Frank (Physics)
• Albert Gapud (Physics)
• Romulus Godang (Physics)
• Armin Straub (Mathematics)

Conference schedule

Talk abstracts

The following is a list of the research presentations with abstracts. Names of presenters are underlined with dots, followed by collaborators and research mentors.

Show/hide abstracts.

1. Suramya Angdembay, Dr. Haiyan Tian (The University of Southern Mississippi)
   Continuous Time Markov Chains for Insider Threat Behavior Modeling
   Abstract. Insider threats often emerge through complex patterns of behavior over time rather than a single action, making detection difficult. This research models user behavior using Continuous Time Markov Chains (CTMC), representing activity as movement through a finite set of human-readable behavioral states over continuous time. Using the CMU CERT Insider Threat Dataset, which contains multi-modal activity logs , the study proposes a hybrid architecture that combines the interpretable local transition structure of a CTMC with the broader contextual and temporal understanding of a Transformer. By utilizing a gated CTMC attention bias, the model injects transition surprisal signals, which measure how unexpected a sequence is, directly into the Transformer's attention mechanism. Validation testing demonstrated that the true CTMC-gated attention consistently outperformed shuffled CTMC controls, proving that the model successfully leverages real behavioral transition structures to improve threat detection and identify anomalous patterns.
2. Nishant Chaudhari, Dr. Ching Shyang Chen (The University of Southern Mississippi)
   Stabilizing RBF Interpolation: A Comparative Analysis of 2D Geometric Profiles and Future Applications in 3D Isosurfaces
   Abstract. The reconstruction of complex, irregular geometries from scattered boundary data is a persistent challenge in computational modeling, where traditional mesh-based methods are computationally expensive. This research explores a meshless alternative by implementing Radial Basis Function (RBF) interpolation to reconstruct implicit surfaces. Because globally supported kernels like the Thin Plate Spline (TPS) frequently result in ill-conditioned linear systems, this study investigates two distinct stabilization techniques: complex regularization (a numerical perturbation) and formal polynomial augmentation.

   Using complex 2D typographic geometries?specifically the letters 'S' and 'Q'?as a fundamental testbed, the methods were evaluated on their numerical stability (RCOND), geometric fidelity, and handling of near-tangent boundary conditions. The results demonstrate that while complex regularization can function adequately for well-behaved geometries, polynomial augmentation provides a mathematically rigorous, scale-independent solution that properly enforces kernel orthogonality and stabilizes the saddle-point system. Ultimately, establishing this optimal balance of conditioning and accuracy in 2D profiles provides a robust, highly efficient framework for scaling these meshless interpolation techniques to complex 3D isosurface reconstructions.

3. Swayam Chaulagain, Dr. Huiqing Zhu (The University of Southern Mississippi)
   A Five Parameter Generalization of Split Normal Distribution
   Abstract. Many statistical models, like normal distribution, assume symmetric data, but real-world observations often exhibit skewness that standard distributions cannot capture properly. The classical split-normal distribution addresses this partially by allowing different variances on each side of the mode, but still forces both sides to share a common center.

   We introduce the Generalized Split-Normal (GSN) distribution, which removes this restriction by allowing the two Gaussian components to have independent means and variances, joined at a flexible threshold point c. We derive the closed-form density, normalization constants, cumulative distribution function, and mean of the distribution, and show that the GSN reduces to the classical split-normal when the two means coincide at c. Parameters are estimated via maximum likelihood estimation (MLE) method and L-BFGS-B (Quasi-Newton) strategy is used for the numerical optimization.

   The GSN is evaluated against the normal, classical split-normal, skew-normal, and a two-component Gaussian mixture model, which has the same number of parameters, using AIC and BIC. Empirical testing on medical insurance charges dataset demonstrate that GSN achieves the best AIC and BIC among all models including the mixture model, while the estimated join point c provides a natural and interpretable threshold separating routine from high-cost claims

4. Prastab Ghimire, Dr. Qingguang Guan (The University of Southern Mississippi)
   Latency-Conditioned Evaluation for EEG Signal Forecasting in Brain-Computer Interfaces
   Abstract. Brain-computer interfaces (BCIs) suffer from processing delays that cause the system to act on outdated neural signals. Forecasting EEG activity ahead of time can compensate for this latency, allowing the system to respond based on predicted brain states. We propose a transformer-based model for short-horizon, multi-channel EEG forecasting designed for BCI latency compensation. The model simultaneously forecasts all 62 EEG channels. A bidirectional encoder processes recent EEG context, and a cross-attention decoder with learnable queries generates each forecast step independently. Reversible instance normalization and a residual connection improve cross-subject generalization and prevent mean collapse. The model produces probabilistic outputs through quantile regression, providing calibrated uncertainty bounds. Training uses a multi-objective loss combining quantile, spectral, temporal, and anti-lag terms, computed only over the useful horizon beyond the latency gap. The model is evaluated on the SEED-IV dataset using a subject-independent protocol where the test subject is entirely unseen during training. Preliminary results show low prediction error relative to typical EEG amplitudes and well-calibrated prediction intervals, suggesting that attention-based forecasting offers a promising direction for reducing the effective latency of real-time brain-computer interfaces.
5. Takeira Joseph, Dr. James Cho (Southeastern Louisiana University)
   Silver Nanoparticles Mediated by Protein-Rich Insect Extracts as Potent Antimicrobials Against Multidrug-Resistant Pathogens
   Abstract. Multidrug-resistant (MDR) pathogens represent a growing global health threat, necessitating the development of alternative antimicrobial strategies. In this study, we report the synthesis of silver nanoparticles (AgNPs) using insect-derived proteins as both reducing and stabilizing agents. These biologically mediated nanoparticles exhibit enhanced stability and surface functionality, enabling improved interactions with bacterial cells. The insect protein extract-based AgNPs demonstrated potent antimicrobial activity against a range of MDR pathogens, including Gram-positive and Gram-negative strains, likely through mechanisms involving membrane disruption, reactive oxygen species generation, and interference with cellular processes. Compared to conventional chemically synthesized AgNPs, the protein-capped nanoparticles showed improved biocompatibility and sustained antimicrobial efficacy. This work highlights the potential of insect protein-mediated nanomaterials as a scalable and eco-friendly platform for combating antibiotic-resistant infections.
6. Nayan Karki, Nicon Pokharel, Dr. Zhifu Xie (The University of Southern Mississippi)
   Searching for Choreographic Orbits in the 5-Body Problem Using the SPBC Method
   Abstract. The Newtonian N-body problem has analytic solutions only for two bodies. For more than three bodies, solutions are chaotic, but rare periodic choreographic orbits exist, where all bodies follow the same path. We apply the Structural Periodic Boundary Conditions (SPBC) method to the planar Newtonian 5-body problem. This variational approach fixes the initial and final configurations using boundary parameters and minimizes the action for all the paths that can exist between the configurations. A two-loop minimization is used where the inner loop solves the boundary value problem for fixed parameters, and the outer loop optimizes the reduced action. We also use the rotational symmetry through a shift condition.
7. Carolyn Lippold, Dr. Romulus Godang (University of South Alabama)
   A measurement of the Branching Fraction of the Upsilon(4S) to B0 and anti-B0 mesons at BABAR
   Abstract. We present a measurement of the branching fraction of Upsilon(4S) decays to B0 and anti-B0 meson pairs at BABAR Experiment. The anti-B0 mesons are partially reconstructed to D*+ lepton anti-neutrino, where the lepton can be either an electron or muon. The D*+ mesons are detected only through the pion daughter of the decay D*+ to D0 pi+. In this talk we discuss an independent measurement of the branching fraction of Upsilon(4S) decays to neutral B pairs.
8. Verity Nwabuisi, Dr. Qingguang Guan (The University of Southern Mississippi)
   Generator-Derived Gate Reduction for Ion Channel Markov Models
   Abstract. Ion-channel kinetics are often modeled by continuous-time Markov chains, where state probabilities evolve across many hidden conformational states and the experimentally relevant output is the open probability. These models are expressive, but they are also high-dimensional and stiff, which makes them expensive to use inside larger electrophysiology simulations. This work studies model reduction for ion-channel Markov models by extracting a small set of generator-derived quantities that capture equilibrium behavior, relaxation timescales, and leading lag effects in the observable dynamics. These quantities are then used to build reduced gate equations for open probability that remain computationally cheap while staying faithful to the dynamics of the full model. The resulting framework connects multi-state channel models to lower-dimensional gate descriptions and is tested against full Markov-chain simulations for several ion-channel models.
9. Dr. Justin Sanders (University of South Alabama)
   Dimensional Analysis
   Abstract. Dimensional analysis is a technique that focuses on the fundamental relationships between the physical quantities relevant to a problem. It is of great help in reducing the experimental complexity of a system and can give insight into the nature of its behavior. This brief introductory tutorial will give the basic ideas of dimensional analysis and provide several examples of its use in physics and engineering.
10. Daniel P. Smith, Dr. Albert A. Gapud, Dr. Rong Cong, Dr. Arneil P. Reyes (University of South Alabama)
    Two superconducting gaps with quasiparticle-lifetime broadening in single-crystal V₃Si revealed by field and temperature dependence of NMR spin-lattice relaxation rate 1/T₁ in ⁵¹V
    Abstract. The A15 superconductor V₃Si - though the subject of nuclear magnetic resonance (NMR) studies for decades - continues to yield insights into the superconducting transition. The measurement of spin-lattice longitudinal relaxation rate, 1/T₁, is a probe into the energy density of states near the Fermi energy, allowing us to model the superconducting gap. We have found that 1/T₁ could be measured more reliably: (1) with a high-purity, single crystal, which in turn (2) allows for an orientation with field along the [111] direction, thus (3) producing a spectrum with one, central transition that follows a single-exponential relaxation curve. Our model for temperature dependence down to less than 100 mK, under fields up to 15 T, clearly indicates two gaps and a "rounding" of the internal gap edge, attributed to the broadening of quasiparticle lifetimes, following a model originally by Dynes et al. in 1978. This behavior suggests an additional mechanism through which vortex penetration accelerates the closing of the gap as critical temperature or field is approached. Gap width and quasiparticle broadening were also found to have contrasting field dependence for the two gaps, in contrast to results from other recent measurements.
11. Morgan Taylor, Aiden McCain, Andy Fritz, Daniel Smith, Dr. Sarah Allred, Dr. Jeffrey A. Mudrock (University of South Alabama)
    Graph Coloring & Enumerative Chromatic-Choosability
    Abstract. Graph coloring was introduced in the 1850s with the famous Four Color Problem about coloring maps in such a way that any two regions sharing a border receive different colors. In this project, we study a variant of graph coloring called list coloring which was introduced in the 1970s. One famous notion in list coloring that has received significant attention in the literature is chromatic-choosability. We examine an enumerative analog called enumerative chromatic-choosability; furthermore, we prove that certain complete multipartite graphs are weakly enumerative chromatic-choosable.
12. Adrianne Van Duzee, Dr. Romulus Godang (University of South Alabama)
    ROOT Program at the High Energy Physics Research
    Abstract. ROOT is a software framework for data analysis. A powerful tool to cope with the demanding tasks typical of state of the art scientific data analysis. The program is based on the C++ programming language. The main features of ROOT which are relevant for the typical problems of data analysis include input and plotting of data from measurements and fitting of analytical functions. In this talk, we will introduce the ROOT program and its features.
13. Zichuo Wang, Dr. Terrence Tsang, Dr. Kazim Sekeroglu (Southeastern Louisiana University)
    Machine Learning-Based Ground Motion Forecasting For Predictive Seismic Isolation Control In Gravitational Wave Detectors
    Abstract. Seismic noise limits the low-frequency sensitivity and operational duty cycle of ground-based gravitational-wave detectors such as LIGO. Active isolation with feedforward control requires accurate short-horizon prediction of the incoming ground motion. This research evaluates machine-learning predictors embedded in a rolling-horizon Model Predictive Control with Feedforward (MPCFF) framework. Five architectures (LSTM, GRU, CNN-LSTM, DLinear, PatchTST) are benchmarked across hyperparameter optimization, input-output window sensitivity, future-window generalization, and seismic isolation control simulation. Recurrent models achieve the lowest prediction error and the best spectral metrics in the 0.06-0.6 Hz control band. GRU, with 30 s of input context, reaches the highest mean band-limited suppression (BLS) of 97.7 times across four test sets. At the harder 1 s prediction horizon LSTM and GRU maintain 71-79 times BLS while simpler architectures collapse. In the MPCFF simulation, the machine learning predictors improve both low-frequency sensitivity and operational efficiency over the linear filter baseline.
14. Aadarsh Yadav, Dr. Zhifu Xie (The University of Southern Mississippi)
    Nonlinear Dynamics of a Predator-Prey Model with Allee Effect and Cooperative Hunting
    Abstract. This paper investigates the nonlinear dynamics of a predator?prey system that incorporates the Allee effect in the prey population and cooperative hunting behavior in the predator. A mathematical analysis of the system?s equilibrium points is performed using the Jacobian matrix to determine conditions for local stability, including extinction, prey-only survival, and coexistence. Numerical simulations are used to support and illustrate the analytical results across different parameter regimes. The study highlights key phenomena such as the Paradox of Enrichment, where increased resource availability can destabilize the system, and the Allee Rescue effect, where threshold adjustments reduce extinction risk. Additionally, the results show that predator intraspecific competition plays a crucial role in suppressing oscillations and promoting stable coexistence.