Session Index
Click a session title to jump to the abstracts.
Faculty Session 1
3:00--4:00, DHC 102
Speakers: Xuemao Zhang, Melissa Innerst, Michael Yatauro
Faculty Session 2
3:00--4:00, DHC 108
Speakers: Tianyu (Timothy) Zhu, Eirini Kilikian, Chuan Li
Faculty Session 3
3:00--4:00, DHC 110
Speakers: Novi Herawati Bong, Sahana H Balasubramanya
Faculty Session 1
DHC 102
3:00-3:15, Xuemao Zhang (East Stroudsburg University)
A Case Study of Learning to Defer for Financial Fraud Detection Using Logistic Regression
View Abstract
Learning to defer allows a predictive model to selectively defer uncertain cases to human experts, balancing automation and oversight. This presentation reports a case study applying threshold-based and joint predictor–rejector learning-to-defer methods to the FiFAR Financial Fraud Alert Review Dataset using the logistic regression model. Using estimated misclassification and deferral costs, we evaluate performance in terms of accuracy, deferral rate, and expected decision cost. The results illustrate how deferral policies affect the trade-off between automated predictions and human review in financial fraud detection.
Close Abstract3:20-3:35, Melissa Innerst (Dickinson College)
Team Cohesion, Motivation, and Stress Resilience in Collegiate Football Players
View Abstract
Grounded in Self-Determination Theory (SDT), we use statistical methods such as nonparametric mediation analysis to determine whether cohesion, specifically task cohesion, improves intrinsic motivation, and subsequently stress resilience. A sample of Division III collegiate football players completed the Group Environment Questionnaire (GEQ), the intrinsic motivation subscale of the Sports Motivation Scale-6 (SMS), and the Nicholson-McBride Resilience Questionnaire (NMRQ). Results indicated that all variables of interest were positively correlated with each other and that intrinsic motivation fully mediated the relationship between team cohesion and stress resilience, in support of SDT.
Close Abstract3:40-3:55, Michael Yatauro (Penn State Brandywine)
Analysis of Drum Performances Using Modern Technology
View Abstract
Given a drumming composition, we can convert the drum strikes into a sequence of times (in seconds) by using the notes and the beats-per-minute. We call this sequence the exact times of the composition. Of course, human drummers do not play notes consistently at these exact times. In fact, doing so tends to create a sound that is machine-like and less audibly appealing. Instead, human drummers are often said to play with “feeling”. For this project, we are interested in comparing the timing of drum notes played by drummers with the exact times and looking for statistically relevant trends. Examples are presented along with a demonstration of the technology used. In particular, we discuss how we use the Moises application for drum track isolation and Python’s Librosa library for audio analysis. This talk provides an update on progress we have made since presenting these results at the 2026 Joint Mathematics Meeting.
Close AbstractBack to index
Faculty Session 2
DHC 108
3:00-3:15, Tianyu (Timothy) Zhu (Temple University)
Neural Network Surrogate for Flux Balance Analysis of Metabolic Models
View Abstract
In computational biology, a metabolic model can be mathematically formalized as a large-scale, underdetermined flow network, defined by a stoichiometric matrix. Flux Balance Analysis (FBA) is a linear programming (LP) framework used to compute the optimal steady-state flow (or "flux") through this network—maximizing a specific objective function, such as growth rate, subject to mass conservation and capacity constraints. Simulating dynamic physical systems, such as biofilm growth, requires solving this LP iteratively across thousands of time steps, creating a severe computational bottleneck. However, the sequence of LPs in these simulations exhibits highly exploitable structural properties: only a restricted subset of the constraint bounds varies over time, and the optimal solution vectors demonstrate strong temporal continuity between adjacent steps. To capitalize on these properties, we introduce a neural network surrogate. By training on a sensitive subset of input-output data generated by exact LP solvers, we construct a fast, approximate non-linear mapping. This surrogate effectively replaces the traditional FBA optimization step within the simulation pipeline, dramatically reducing computational complexity and execution time while preserving the fidelity of the network flows.
Close Abstract3:20-3:35, Eirini Kilikian (University of Delaware)
Agent-Based Modeling of Idiopathic Lung Fibrosis and Mechanistic Treatments
View Abstract
This agent-based model simulates interactions between fibroblast and myofibroblast cells during idiopathic pulmonary fibrosis (IPF) in alveolar tissue microenvironments. By integrating computational modeling of IPF and therapeutics, this research aims to improve understanding of fibrosis progression and assess the efficacy of novel and existing treatments targeting different mechanisms to inform decision-making for IPF treatment.
Close Abstract3:40-3:55, Chuan Li (West Chester University of PA)
A Numerical Study of Heat Dissipation in Human Tumor Anatomies during Magnetic Fluid Hyperthermia
View Abstract
Magnetic fluid hyperthermia (MFH) is an effective clinical cancer therapy in which magnetic nanoparticles are delivered into tumors and heated using an alternating magnetic field. This talk presents a recently developed image-based computational framework for simulating MFH in irregular human tumor anatomies reconstructed from CT-derived mask images. The proposed model incorporates Pennes’ bioheat equation and Rosensweig magnetic heating, solved by a novel Augmented Matched Interface and Boundary (AMIB) method for temperature evolution. In addition, the Arrhenius thermal damage equation is employed to predict irreversible tumor tissue damage in the tumor. Numerical simulations on representative liver tumor geometries are presented to evaluate thermal damage and treatment efficacy. The results demonstrate that the AMIB-based MFH simulation framework is a practical and effective tool for patient-specific treatment planning and optimization.
Close AbstractBack to index
Faculty Session 3
DHC 110
3:00-3:15, Novi Herawati Bong (University of Delaware)
Mentoring Graduate Student Instructors in Mathematics: Reflections and Lessons Learned
View Abstract
In Fall 2022, I began mentoring graduate students in mathematics at the University of Delaware as they developed their identities as instructors. This role has involved designing a 3-credit teaching course and observing their classroom practices over multiple semesters. In this presentation, I reflect on over three years of experience, sharing successes, challenges, and practical approaches to mentoring that support graduate students in becoming more effective and confident mathematics educators.
Close Abstract3:20-3:35, Sahana H Balasubramanya (Lafayette College)
Extending acylindricity to higher rank
View Abstract
I will present a new notion of non-positively curved groups: the collection of discrete countable groups acting (AU-)acylindrically on finite products of hyperbolic spaces. This work (joint wt T.Fernos) is inspired by the classical theory of $S$-arithmetic lattices and that of acylindrically hyperbolic groups. I will focus on the motivation for studying these groups in this talk.
Close AbstractBack to index