Abstracts of Oral and Posters

ICM 2026 Satellite Conference - Complete Professional Program Schedule 

Abstracts of Oral and Posters 

Jackknife Empirical Likelihood Method for Missing Data and Measurement Error 

Hakeem Adekunle 

Georgia State University 

Measurement error, often treated as a missing data problem, poses major challenges in health and social science  research by introducing bias and unreliable inferences. This thesis applies the Bayesian Jackknife Empirical  Likelihood (BJEL) method to handle measurement error using jackknife pseudo-values based on semiparametric  fractional imputation (SFI), propensity score (PS), and doubly robust (DR) estimators. Through simulation studies,  BJEL was compared with traditional methods, including Jackknife Normal Approximation (JNA), Jackknife Empirical  Likelihood (JEL), and Bootstrap (BS). Results showed BJEL outperformed the alternatives in terms of credible  interval and average length, particularly with small sample sizes. The DR estimator showed notable robustness,  maintaining reliable coverage even under model misspecification. The method was further applied to 2015–2016  NHANES data to evaluate the association between perfluoroalkyl acid (PFA) exposure and impaired kidney  function.  

Impact of Drought Indices on Maize Yield in Northern Nigeria Using Multiple Linear Regression and Spatial  Analysis 

Grace Adenuga 

University of Nevada Las Vegas 

Drought is a key abiotic stress affecting maize yield and production in Sub Saharan Africa contributing between  44% to 58% grain yield decline in West and Central Africa. For the detection, classification, and control of drought  conditions, drought indices are used. This paper presents the application of a multiple linear regression model and  spatial distribution to assess the performance of drought indices on maize production in the Northern part of  Nigeria. In this research, observed annual data of drought indices, RDI and the Palmer drought indices which  includes SCPDSI, SCPHDI and SCWLPM, maize yield (measured in tonnes) in Northern states of Nigeria were  obtained from 1993 to 2018. The multiple linear regression was carried out at different training sets: 70%, 80% and  90%. Results from the multiple linear regression showed that in the North-Central states, FCT has the lowest MSE  (0.7788234) at 90% training level. In North-Eastern states, Borno state has the lowest MSE (0.7240276) at 80%  training sets. In Northwestern states, Kebbi state has the lowest MSE (0.8029484) at 70% training set. From the  spatial distribution, Kwara state has the lowest maize yield among the North central states. Yobe state has the  lowest maize yield among the Northeast states. And Sokoto state has the lowest maize yield among the Northwest  states. Overall, Yobe state has the lowest maize yield in the Northern states. The findings demonstrate that  drought indices significantly influence maize yield variability and that regional differences are important for  predictive modeling. This study highlights the usefulness of multiple linear regression combined with spatial  analysis for understanding drought impacts on agricultural productivity and supporting data-driven agricultural  planning in drought-prone regions. 

Comparing Finite-Difference and Spectral Element Methods for Nonhydrostatics Atmospheric Dynamics:  Benchmark Results from WRF and Jexpresso 

Olayemi Adeyemi 

Boise State University 

Accurate numerical modeling of atmospheric dynamics is critical for weather prediction and climate simulation.  This talk presents a comparison of two discretization approaches: the finite-difference method in the Weather  Research and Forecasting (WRF) model, and the continuous Galerkin spectral element method in Jexpresso. Using  nonhydrostatic mountain wave benchmarks; including a linear small-amplitude case and the nonlinear Jang &  Hong case, we evaluate how each method captures wave dynamics and how numerical dissipation parameters  affect solution quality. Results show that both methods reproduce benchmark behavior, with key differences in  dissipation tuning. This work informs the broader question of whether high-order spectral element methods are  viable for next-generation atmospheric modeling.

ICM 2026 Satellite Conference - Complete Professional Program Schedule 

A Novel Smooth Fixed-Time Controller Design for Chaotic Spacecraft Formation Synchronization. Israr Ahmad 

Coppin State University, Baltimore, Maryland 

This study addresses two key gaps in the research on chaos synchronization (i) It explores fixed-time leader– follower spacecraft formation synchronization, an area that has not been previously explored (ii) The study also  introduces a new fixed-time feedback control law that overcomes the limitations of current methods. The  proposed law enables faster, smoother, and more robust synchronization, ensuring guaranteed settling time and  minimizing state-variable fluctuations during reconfiguration. Lyapunov-based analysis confirms that state errors  converge to zero within a set time, regardless of initial conditions. Numerical simulations involving one leader and  three followers with varying parameters and attitudes demonstrate the method's effectiveness. Comparative  analysis highlights the novelty and superior performance of the proposed control law, supporting its potential for  high-precision, resilient spacecraft formations. 

Recovering Constant and Time-Varying Parameters in Epidemiological Systems using data assimilation Muhammad J. Ahmad 

University of Maryland Baltimore County 

Estimating parameters in epidemiological models is often difficult by the fact that some parameters remain  constant while others evolve in time. Many existing approaches treat these two settings separately, which can limit  their applicability when both types of parameters appear in the same model. In this work, we develop a data  assimilation-based optimization framework for estimating constant and time-dependent parameters from partial  observations of the state. The method incorporates observational feedback into the model dynamics and  formulates parameter recovery as a local-in-time optimization problem. This allows time-varying parameters to be  reconstructed over the full observation interval while constant parameters are estimated within the same  computational framework. We first test the approach on a Susceptible–Infected–Susceptible (SIS) model with a  time-dependent transmission rate. We then apply the method to a Susceptible–Infected–Vaccinated–Recovered– Deceased (SIVRD) influenza model using CDC incidence data. The results demonstrate stable parameter  reconstructions that are robust to observational noise, highlighting the effectiveness of the proposed framework  for inverse problems in nonlinear dynamical systems. 

Exploring Beta Cell Coupling Conditions: A Genetic Algorithm Approach to Critical Synchronization Transitions Zainab Almutawa 

University of Maryland Baltimore County 

Pancreatic beta cells regulate blood glucose levels by secreting insulin in a coordinated manner, a process that  critically depends on electrical coupling through gap junctions. Disruptions in this coupling - as well as variations in  cellular excitability—can disturb synchronized insulin release and may contribute to the development of diabetes.  In this study, we explore how changes in parameters governing gap junction connectivity and cell excitability shape  network dynamics using a computational model of beta cells. We further establish a necessary condition for  synchronization on coupling using the Master Stability Function. A genetic algorithm was employed to scan the  multidimensional parameter space, identifying candidate parameter sets that achieve robust synchronization  under normal conditions yet lead to complete desynchronization when a specific cell is silenced via chloride  current activation. In parallel, a neural network classifier was developed to distinguish between candidate sets  exhibiting these distinct behaviors, confirming that silencing a critical cell can trigger a switch-like transition in  network dynamics. Statistical analyses highlight the influence of individual parameters on synchronization. These  findings advance our understanding of beta cell network behavior and provide new insights into the mechanisms  underlying coordinated insulin secretion and its dysregulation in diabetes.

ICM 2026 Satellite Conference - Complete Professional Program Schedule 

A Mathematical Model of COVID-19 Transmission Dynamics with Application 

Aminah Alrehaili 

Morgan State University 

In this poster, a preliminary mathematical model of COVID-19 transmission dynamics with application to the U.S.  population will be presented. The mathematical model consists of a system of ordinary differential equations that  describes the spread of the disease within a population. The basic reproduction number and the effective  reproduction number were derived and analyzed to assess disease transmission. The model’s numerical  simulations were also conducted and will be presented. 

Embedded implicit-explicit strong stability preserving Runge-Kutta methods 

Sylvia Amihere 

University of Maryland, Baltimore County 

Strong Stability Preserving (SSP) Runge-Kutta methods are time integration schemes designed to preserve  monotonicity and nonlinear stability properties of spatial discretization methods when solving time-dependent  differential equations. In recent times, new classes of problems have emerged which can be modeled by time dependent equations that involve both stiff and non-stiff terms, where the non-stiff terms are solved explicitly and  stiff terms are treated implicitly. This has motivated the development of Implicit-Explicit (ImEx) SSP methods  designed for fixed time stepping. However, this may not be ideal for problems with rapidly changing dynamics. To  address this limitation, we introduce a new class of ImEx SSP methods that employ adaptive time stepping to  adjust step sizes while preserving stability properties. Numerical experiments are conducted to analyze the  efficiency and robustness of these new methods. 

Modeling the Progression of Retinal Detachment (RD): Exploration of Machine Learning Techniques

William Annan 

Texas Christian University 

Retinal detachment (RD) is the separation of the neural layer (NL) from the retinal pigmented epithelium (RPE). In  a healthy eye, the retina, consisting of the neural layer and the retinal pigmented epithelium, is firmly attached to  the eyewall. The separation of these layers prevents the supply of nutrients to the cells within the neural layer of  the retina, thereby disrupting the normal function of the photoreceptor cells, including the daily renewal process  involving disc addition and removal to prevent the buildup of toxins. If retinal detachment is not detected and  treated promptly, it can result in permanent vision loss. However, the rate at which retinal detachment progresses  and the effective timeframe for optimal treatment are not well understood. In this work, we develop a two dimensional fluid--structure interaction model to investigate the rate of disease progression and the factors that  most strongly influence it. Due to the complexity and computational cost of the model, we are also investigating  machine learning techniques that may help extend the model to three dimensions, incorporate the geometry of  the eyeball, reduce computational cost, and improve model performance. 

Exponential time differencing schemes with rational approximations for solving semi linear evolution equations Emmanuel Asante-Asamai 

Clarkson University, Potsdam, NY 

Time evolution of semi linear partial differential equations (PDE) face significant numerical challenges due to  stiffness, high spatial dimensions, non-smooth initial and boundary data and coupled nonlinear source terms.  Explicit forms of traditional time stepping methods such as Runge-Kutta, Theta-methods or Multistep methods  have relatively small regions of absolute stability that limit their applications to stiff PDE. Their implicit counter  parts, which tend to have larger stability regions, require the use of interactive solvers to handle coupled nonlinear  source terms, resulting in a significantly higher runtime. Whereas efforts to increase the stability region while  reducing runtime using implicit-explicit methods (IMEX) have been largely successful, they require the generation  of a sufficiently accurate history of initial data and are often ineffective in damping out spurious oscillations that  emerge due to non-smooth initial and/or boundary conditions. In this talk, I will present new exponential time  differencing schemes which are A-stable, and so do not impose a stability restriction on time step, and do not  require iterative solvers to handle nonlinear source terms. These schemes are also L-stable, making them effective  in damping spurious oscillations, and use a dimensional splitting technique to reduce the computational cost of  solving multidimensional problems. The advantage of these schemes over traditional methods will be illustrated  through several nonlinear reaction-diffusion equations. 

Polynomial-Preserving Linear Filters: A Comprehensive Derivation of the Spencer Filter and Its Unification with  the Savitzky-Golay Smoothing Filter 

Kaosarat B. Azeez 

Illinois State University 

Polynomial-preserving linear filters play a central role in smoothing and trend extraction for noisy time series,  where the goal is to reduce short-run variation without distorting underlying low-degree structure. This work  introduces a unified framework for such filters based on local polynomial approximation and least-squares  estimation, showing that the resulting estimators can be expressed as symmetric linear filters whose weights  satisfy moment conditions ensuring exact polynomial reproduction. Within this framework, explicit coefficient  vectors are derived for classical smoothing methods, including the Savitzky-Golay filters SG (5,2) and SG (7,3), and  a central contribution of this work is a complete, self-contained derivation of the Spencer 15-point smoothing  filter, whose full construction is rarely presented in literature. Analytical verification of polynomial reproduction  properties demonstrates that, despite their distinct constructions, both Savitzky-Golay and Spencer filters satisfy  the same moment-condition constraints, providing a unified interpretation of these methods. Numerical  experiments using synthetic and real-world time series illustrate the trade-off between smoothness and  responsiveness, clarifying how polynomial reproduction, symmetry, and window length jointly determine filter  performance. 

Small diameter properties in Banach Spaces 

Sudeshna Basu 

Loyola University 

The geometry of Banach space is an area of research which characterizes the topological and measure theoretic  concepts in Banach spaces in terms of geometric structure of the space. In this work we study three different  versions of small diameter properties of the unit ball in a Banach space and its dual. The related concepts for all  closed bounded convex sets of a Banach space were initiated developed and extensively studied in the context of  Radon Nikodym Property and Krein Milman Property in [1]and developed subsequently. We prove that all these  properties are stable under lp sum for 1 ≤ p ≤ ∞, c0 sum and Lebesgue Bochner spaces. We show that these are  three space properties under certain conditions on the quotient space. We also study these properties in ideals of  Banach spaces. This is based on several papers jointly written with my graduate student, Susmita  Seal([2],[3][4],[5]). The only prerequisite for this talk is the statements of Hahn Banach Theorem. 

Simulated signaling of DNA origami nanostructures for biosensor development 

Alli Carson 

National Institute of Standards and Technology 

Biological field effect transistors (Bio-FETs) have shown great promise in revolutionizing diagnostic testing,  enabling the development of inexpensive and portable biosensors with a very low limit of detection that are  capable of pro point-of-care diagnostics within minutes. A similar biosensor employing DNA origami  nanostructures has shown unique potential, as the nanostructures’ strong negative charge and low permittivity is  thought to significantly affect the electric field within the device, resulting in a robust signal for a positive  diagnosis. This talk develops a mathematical model and simulation of the electrochemical process, using a  modified Poisson-Boltzmann equation to describe the behavior of the electric potential about a single  nanostructure. Nonsmooth and smooth representations of the nanostructure are motivated and analyzed. Three  different nanostructure designs are studied and their capacity for signaling compared with suggestions on test  design.

The Syntax of Equity: Quantifying LLM Grading Bias via Formal Grammar Modeling and Automata Theory

Chidubem Chidebelu 

Morgan State University, Computer Science Department 

As Large Language Models (LLMs) are increasingly integrated into academic grading pipelines, a critical risk emerges: the  conflation of linguistic variation with cognitive or intellectual deficiency. This research operationalizes the 'Fairness Gap' by  treating student writing not as a monolithic entity, but as a membership problem within distinct formal languages. Specifically,  we define two Context-Free Grammars (CFGs): G_SE (Standard Academic English) and G_VAR (a variant dialect, such as AAVE or  L2 English transfer patterns). While modern LLMs are often evaluated on semantic benchmarks, this study investigates their  behavior as restricted Pushdown Automata (PDA). By generating a 'Ground Truth' dataset of strings that satisfy the production rules of P_VAR but are rejected by P_SE, we isolate syntactic variation from logical errors. We evaluate open-source models— 

Llama-3 (8B) and Mistral-7B—to determine if they penalize strings belonging to L(G_VAR) despite equivalent semantic value to  strings in L(G_SE). The study provides a Python-based implementation framework to categorize essay components before LLM  processing, shifting the discourse from qualitative 'fairness' to the formal validation of diverse linguistic production rules. 

Extending Inexact Anderson Acceleration to Asynchronous Regimes 

Evans Coleman 

University of Mary Washington 

Fixed-point iterations are ubiquitous across scientific computing, but their traditional synchronous execution faces scaling  bottlenecks on increasingly heterogeneous hardware. Asynchronous models have emerged to eliminate these rigid bottlenecks  by allowing processors to update independently using delayed data, and while this maximizes throughput, the resulting state  inconsistencies can severely degrade baseline convergence. Anderson Acceleration is a highly effective convergence accelerator  for synchronous methods, but its viability under asynchronous noise remains largely unexplored. This talk examines how  asynchronous artifacts interact with the Anderson subspace. Because asynchronous iterates are composite states formed from  partial worker returns, the residuals fed into the Anderson history are continually evaluated at inconsistent points, challenging  the foundational assumptions of the accelerator. We demonstrate that the success of this acceleration subspace depends  fundamentally on the operator's coupling density. For densely coupled operators, each partial update carries global  information; staleness acts merely as a bounded perturbation to the map evaluation, allowing Anderson Acceleration to  successfully correct the asynchronous bias. Conversely, for sparsely coupled maps, stale returns encode only local boundary  information. When assembled into the global iterate, they act as iterate-level corruption, filling the subspace with artifacts  rather than the global error modes necessary for extrapolation. Building on prior theory for inexact function evaluations, we will outline how the interaction between coupling density and extrapolation firing frequency dictates the quality of the  accelerated output, establishing the strict scheduling controls required to keep the history well-conditioned and restore  convergence. 

Locking-Free Solvers for Time-Harmonic Biot Poroelasticity: Finite Elements versus Physics-Informed Neural  Networks in 2D and 3D 

Stanley Diala 

Morgan State University 

The time-harmonic Biot model describes coupled solid–fluid interaction in saturated porous media and underpins Magnetic  Resonance Elastography (MRE), a non-invasive technique for reconstructing tissue shear modulus, hydraulic conductivity, and  interstitial pressure in organs including the liver, lung, and brain. This work systematically compares finite element methods  (FEM) and physics-informed neural networks (PINNs) for the three-field time-harmonic Biot system, employing a locking free  total-pressure formulation φ = αp − λ∇·u that guarantees well-posedness in the nearly incompressible regime. Three algorithms  are evaluated: a coupled monolithic scheme (CwtNoSta) and two iterative decoupled schemes—pressure-first (IDA-I) and  displacement-first (IDA-II)—using manufactured solutions on the unit square and unit cube (up to 1.79M degrees of freedom),  with sweeps over Poisson ratio, permeability, and frequency. FEM achieves 10⁴–10⁶× higher accuracy than PINNs at 100–400×  lower cost. Both iterative schemes match monolithic accuracy while delivering wall-time speedups of approximately 2× in 2D  and up to 19% in 3D. The outer-iteration count is mesh-independent—six in 2D, five in 3D—consistent with theoretical  contraction estimates governed solely by physical parameters. After one initial factorization, subsequent iterations reduce to  inexpensive back-substitutions, conferring both computational and memory advantages over the full coupled system. For  repeated forward solves on a fixed mesh, each additional evaluation with IDA-I or IDA-II requires only cached back-substitutions  and a scalar pressure solve, avoiding full re-factorization and yielding substantial gains in execution efficiency as the number of  solves grows. The total-pressure formulation is validated as locking-free and parameter-robust across both dimensions and  paradigms.

Variational Implicit Solvation model with Size-Modified Poisson Boltzmann and Diffuse interface

Matthias Dogbatsey 

The University of Alabama 

Variational implicit solvation models (VISMs) have been widely used in the solvation analysis of chemical and  biological systems at molecular level. Central to the construction of a VSIM is an interface separating the solute  and the solvent. In this work, A novel diffuse-interface VISM will be developed, based on a physical diffuse  interface definition that is rigorously derived from statistical mechanics. The proposed model can capture the  ensemble average solvation energy (EASE), which is the experimentally observable energy during the solvation  process. Further, the non-ideal solvent properties such as finite ion size effect have been considered to construct a  more physical solvation model. 

Computational Complexity of Turing Machine Simulations in Modern Computing 

Joseph Egbuanran 

Morgan State University Computer Science 

High-Dimensional Regularization of the Spatial Sign Covariance Matrix for Robust Shape Estimation

Jonas Elmerraji 

Johns Hopkins University 

High-dimensional covariance matrices are of fundamental interest in real-world problems spanning fields such as  finance, image processing, and genomics. When the number of dimensions is comparable to the number of  observations, however, classical frameworks for evaluating covariance and shape matrix estimators no longer  apply. This project studies the estimation of the shape matrix, which encodes covariance structure up to scale, in a  high-dimensional generalized elliptical model. We compare the spatial sign covariance matrix (SSCM), the sample  covariance matrix of Euclidean-normalized observations, with standard alternatives typically favored in shape  matrix estimation, including unbiased estimators and robust alternatives such as Tyler’s M-estimator (TME). While  the SSCM is known to be statistically inconsistent under classical fixed-dimension asymptotic theory, we find that  the SSCM can outperform standard alternatives in high-dimensional regimes. Using recent tools from random  matrix theory, we show that, under high-dimensional asymptotics, the SSCM achieves lower expected estimation  error than TME in this model. We further show that the SSCM has lower risk than any unbiased estimator of  Gaussian-scale mixture shape matrices under squared-error loss in this regime, a comparison that is especially  relevant for likelihood-based estimators, which are often asymptotically unbiased. In both cases, this improvement  comes from a second-order implicit regularization effect that is not predicted by classical fixed-dimension  asymptotics. Simulations support the theory: predicted high-dimensional errors remain strikingly accurate even in  moderate dimensions, and SSCM-based estimates can be computed at vastly lower cost than TME-based  estimates. We further extend the method to covariance and dispersion matrix estimation for elliptical  distributions, where the resulting estimators empirically outperform standard benchmarks in high-dimensional  settings. These results suggest further opportunities to exploit second-order effects in high-dimensional robust  estimation. 

Exponential decay of a swelling porous system with Lord-Shulman type thermoelasticity

Cyril Enyi 

Mount Vernon Nazarene University 

We present an observed exponential decay rate for a swelling porous system under the thermal effect produced  by Lord-Shulman thermoelastic theory. Well-posedness result is established using the Lumer-Phillips corollary  alongside the Hille-Yoside theorem. The search for stability is explored via the energy/multiplier method in a  manner that precludes the imposition of an equal wave speed of propagation. Numerical computations are  performed to validate the theoretical analysis.

Self-Adaptive Inertial Algorithms for Equilibrium Problems with Applications to Heart Disease Classification

Amara Eze 

Morgan State University 

Heart disease continues to be one of the major causes of death globally, creating serious challenges for public  health systems. Accurate early diagnosis and reliable prediction techniques are essential for reducing its impact,  since untreated conditions may result in severe complications affecting important organs such as the heart,  kidneys, and nervous system. In this study, we introduce a new inertial Mann-type Tseng extragradient algorithm  with self-adaptive step sizes for solving Split Pseudomonotone Equilibrium Problems involving multiple output sets  in real Hilbert spaces. In contrast to many existing methods that rely on strong assumptions such as co-coercivity  and Lipschitz continuity, the proposed approach requires only monotonicity and uniform continuity of the  associated single-valued operators, thereby increasing its applicability to practical real-world problems. Moreover,  the proposed scheme does not require prior estimation of the operator norm, making it more flexible and  computationally efficient in applications. Under suitable assumptions on the control parameters, we establish  strong convergence of the generated iterative sequence to the minimum-norm solution of the considered Split  Pseudomonotone Equilibrium Problem. Finally, we show the effectiveness of the proposed method through  experiments on heart disease prediction in machine learning and compare its performance with existing  algorithms to demonstrate its superiority and practical relevance.  

Retrieval-Augmented Machine Learning for Accelerating Scientific Computation in Data-Driven Systems

Michael Ezeigbo 

Morgan State University 

Modern scientific computing increasingly integrates numerical solvers with machine learning to accelerate  simulation, modeling, and data-driven analysis. However, repeated evaluations of discretized operators, partial  differential equation solvers, and iterative linear and nonlinear systems remain computationally expensive in large scale scientific workflows. This work introduces a retrieval-augmented machine learning framework that combines  embedding-based memory retrieval with surrogate modeling to accelerate scientific computation. The system  maintains a structured memory of prior computational states, including intermediate solutions, boundary  conditions, and operator evaluations. During inference, relevant historical computational contexts are retrieved  and used to guide approximation of expensive numerical procedures, enabling reuse of prior computation across  iterative tasks and reducing redundant evaluation of numerical processes. We evaluate the framework on  representative scientific prediction tasks involving simulation-based inference and numerical approximation,  comparing it against standard surrogate modeling and machine learning baselines. Results show consistent  reductions in computational cost and runtime while maintaining bounded approximation error across diverse test  cases. The proposed framework supports scientific computing pipelines involving PDE solvers, discretized  operators, and coupled dynamical systems, enabling efficient reuse of computational history. This work  demonstrates that retrieval-augmented learning can transform expensive numerical workflows into memory aware computational systems that scale efficiently without sacrificing accuracy 

Data-driven Integrated Framework for Modeling Protein Interaction Networks 

Elisha Erzoah 

George Mason University 

Mitogen-Activated Protein Kinase (MAPK) pathway is a signaling pathway playing critical part in cell proliferation,  survival and death processes. MAPK modeling faces challenges associated with the network complexity and lack of  sufficient experimental data. Moreover, there is a danger of over-relying on modeling assumptions when picking  critical features of the signaling pathway, which complicates the study of homeostasis patterns. We present a new  data-driven approach to modeling this pathway integrating kinetic and graph theoretical methodology. By applying  a combination of graph centrality measures to the curated larger protein network, we can automatically identify  which proteins and pathways are most dominant, justifying the selection of nodes in the reduced low-dimensional  representation. In addition, entropy-based measures encoding signal correlations are used to investigate changes  in information flow in mutated cell lines compared to the normal cell lines. This framework combining network  analysis, dynamical systems, and parameter estimation has a potential to detect changes in pathway organization  and signaling dynamics in complex biological systems important for biomedical applications.

Existence and uniqueness of solutions to impulsive fractional differential equations via the deformable  derivative 

Mesfin Etefa 

Bowie State University 

This talk investigates the existence and uniqueness of solutions for a class of impulsive fractional differential  equations involving a recently introduced fractional operator known as the deformable derivative. The model  incorporates instantaneous impulses, allowing the description of dynamical processes that undergo abrupt  changes of state at specified moments in time. Such impulsive phenomena arise naturally in many applications,  including biological systems, mechanical processes, control theory, and fluid dynamics, where both memory effects  and sudden perturbations play an important role. The presentation begins with a brief introduction to fractional  differential equations and the motivation for using the deformable derivative. We then formulate the impulsive  problem and establish sufficient conditions for the existence and uniqueness of solutions. The analysis relies on  several classical tools from nonlinear functional analysis, including the Banach contraction principle, Schaefer’s  fixed-point theorem, and the Leray–Schauder alternative. To demonstrate the applicability of the abstract results,  an illustrative example is presented. 

SQP Analysis with Constrained Activation Regimes in Multi-objective Optimal Control

Romario Gildas Foko Tiomela 

Morgan State University 

This study analyzes a multi-objective optimal control problem through a Sequential Quadratic Programming (SQP)  framework combined with an ε-constraint formulation. The Karush-Kuhn-Tucker (KKT) conditions associated with  the resulting quadratic subproblem lead to four distinct regimes, corresponding to all possible combinations of  active and inactive inequality constraints. For each regime, explicit a priori estimates for the SQP control update  are derived, providing practical criteria for convergence analysis and step-size assessment. Numerical simulations  based on a compartmental epidemic model validate these theoretical estimates. 

Synchronization and Coupling Dynamics in Multi-Population Epidemic Systems 

Sunday O. Gbodogbe 

Indiana University Indianapolis 

The increasing connectivity between populations through mobility and travel can produce complex  synchronization patterns in infectious disease outbreaks across regions. Understanding these dynamics is  important for epidemic forecasting, intervention planning, and assessing the risk of coordinated outbreak waves.  In this work, we investigate synchronization phenomena in coupled epidemic systems with heterogeneous  transmission dynamics and mobility-driven interactions between populations. Using a multi-population  compartmental modeling framework, we analyze how coupling strength, parameter heterogeneity, and temporal  forcing influence the transition between synchronized and desynchronized epidemic behavior. Computational  simulations reveal the emergence of phase-locked outbreak dynamics, delayed synchronization, and partial  synchrony under varying interaction structures. We further examine how differences in local epidemiological  conditions affect outbreak timing and epidemic coherence across populations. The results demonstrate how inter regional coupling can shape large-scale epidemic patterns and provide insight into the role of synchronization in  infectious disease spread. This work contributes to the growing interface between mathematical epidemiology,  dynamical systems, and computational modeling of complex biological systems. 

Numerical Analysis of a Coupled 3D-1D Transport Problem 

Uzochi Gideon 

Rice University 

A finite element solution coupled with an interior penalty discontinuous Galerkin solution are defined for the  approximation of the coupled 3D-1D solute transport problem. Under sufficient regularity for the weak solutions,  optimal error bounds are derived for the 3D concentration and 1D concentration, that are optimal with respect to  the time step size and the mesh sizes. Numerical results verify the theoretical results.

Rendezvous Search on a Sphere: The Astronaut Problem 

Drew Henrichsen 

Johns Hopkins University 

In a rendezvous search problem, the goal is to find an optimal strategy for two agents in a specified region to use in  order to meet in the minimum expected time. The astronaut problem considers two agents on a sphere  representing a featureless planet. The agents can see everything within a detection radius r and have a maximum  allowed speed of movement. Although the problem is decades old, there are no published results. We consider  multiple strategies, providing both theoretical bounds and Monte Carlo simulation estimates. Some of the  strategies are iterative, restarting if the agents have not met, so they involve geometric waiting time distributions.  Upper bounds that vary inversely with r are obtained in both the symmetric and asymmetric settings, providing the  first recorded finite bounds for the problem. In addition, a general bound is given for a large class of higher  dimensional search regions. 

Robust Impulse Control in Financial Markets with Drift Uncertainty and Transaction Costs

Temitope C. Iroko 

University of Wisconsin-Milwaukee 

Transaction costs make continuous trading impractical in many financial markets, motivating impulse control  models in which investors adjust their positions only at discrete intervention times. We study an infinite-horizon  impulse control problem for a risky holding modeled by a geometric Brownian motion, where each intervention  reduces the holding and incurs a fixed transaction cost. In the classical formulation, the asset drift is assumed to be  known, and the value of a threshold policy admits a resolvent representation involving a particular solution and a  homogeneous fundamental solution. This reduces the optimization over threshold policies to the maximization of  a scalar function of the intervention levels. We extend this framework to a robust setting in which the investor is  uncertain about the drift of the asset. Drift misspecification is incorporated through a worst-case distortion of the  model, penalized by a Maenhout-type homothetic penalty. The resulting robust Hamilton–Jacobi–Bellman quasi variational inequality contains a nonlinear gradient term, so the classical superposition argument and explicit  threshold payoff decomposition no longer apply. We retain the threshold-policy structure and characterize the  robust value function through a nonlinear free-boundary problem. A logarithmic transformation is used to obtain a  boundary-value formulation, with value-matching and smooth-pasting conditions determining the intervention  thresholds. Numerical results illustrate the robust value function, the optimal intervention thresholds, and the  worst-case drift distortion. 

Performance Analysis of Cloud-Based Machine Learning for Scientific Computing Applications

Serenity Jackson 

Morgan State University 

Cloud computing has emerged as a powerful platform for scientific computing by providing scalable, accessible,  and cost-effective computational resources for data analysis and machine learning applications. This project  investigates the use of cloud computing for scientific computing through the implementation of Python-based  machine learning techniques within the Google Collab environment. Scientific computing libraries, including  NumPy, Pandas, Matplotlib, and Scikit-learn, were utilized to preprocess, analyze, and visualize healthcare-related  data. A diabetes prediction dataset was employed to develop and evaluate a machine learning classification  model. Data preprocessing and exploratory analysis were performed using Python scientific computing methods to  improve data quality and model performance. A Logistic Regression algorithm was implemented to predict patient  outcomes based on key medical attributes such as glucose level, blood pressure, body mass index (BMI), insulin  level, and age. The model was trained and tested using the cloud-based computational infrastructure provided by  Google Collab. Experimental results demonstrate that cloud computing platforms can efficiently support scientific computing and machine learning workflows while providing high accessibility and reduced infrastructure costs. The  developed model achieved strong predictive accuracy, highlighting the effectiveness of cloud-based resources for  healthcare analytics and large-scale data processing tasks. This study emphasizes the increasing significance of  cloud-enabled machine learning technologies in advancing scientific computing, medical data analysis, and modern  research applications.

Efficient Stochastic Search for D-Optimal Design 

Yunze Ji 

Johns Hopkins University 

Finding D-optimal experimental designs for complex nonlinear models is highly computationally intensive,  especially in continuous, high-dimensional spaces. Standard grid-based methods become practically infeasible as  dimensions grow, while gradient-based techniques are limited by the high cost and difficulty of evaluating the  determinant of a large Fisher Information Matrix (FIM). To overcome these bottlenecks, we introduce a novel  framework that couples Simultaneous Perturbation Stochastic Approximation with Stochastic Lanczos Quadrature  (SPSA-SLQ). Because our approach avoids forming the FIM explicitly or factoring its determinant directly, it  searches continuous parameter spaces efficiently without relying on grids. We introduce three variants designed  for different problem settings: (1) SPSA-Slogdet, a precise baseline using exact determinant calculations for lower dimensional problems; (2) Cyclic SPSA-SLQ, which uses block coordinate descent to alternate between updating  support points and weights for faster convergence; and (3) Averaged SPSA-SLQ, which averages independent  gradient estimates to reduce search noise and improve final accuracy. We evaluate this framework on 12  benchmark problems, including complex generalized linear models and multinomial logistic regression models with  up to 22 parameters. Our results show that the framework achieves D-efficiency scores that match or outperform  state-of-the-art evolutionary algorithms like LSHADE and JADE. Using a two-stage pruning strategy during and after  optimization, the method removes unnecessary support points to produce sparse, practical designs. It also  successfully finds minimal, valid support sets in structurally rank-deficient systems. Most importantly, the SPSA  optimization engine maintains a fixed evaluation cost per iteration regardless of the parameter dimension, offering  a highly scalable approach for optimal experimental design in complex nonlinear models. 

Approximate Solution of an Epidemic SIR Model via Laplace Transform and q -Homotopy Analysis Method

Lubna Kadhim 

Morgan State University 

Fractional differential equations have emerged as powerful tools for modeling complex dynamical systems with  memory and hereditary properties. In this study, we employ fractional-order derivatives to analyze dynamical  models that capture the physiological mechanisms underlying multiple diseases. Although classical models are  typically formulated using ordinary and partial differential equations, we introduce a generalized framework based  on fractional differential equations of order 0 < n ≤ 1, thus enhancing modeling accuracy and flexibility. This  approach contributes to a field that has evolved steadily since its early foundations in 1730. 

Population Movement and Endemic Persistence in SIS Epidemic Network Models 

Ukandu Kingsley 

University of Nevada, Las Vegas 

Understanding the spatial structure of endemic equilibria in epidemic models is central to characterizing the long term dynamics of infectious diseases and developing effective control strategies. In the literature on susceptible infected-susceptible epidemic patch models, it is known that, under certain conditions, an infectious disease can  be eradicated when the basic reproduction number is less than one. In this talk, we present new results showing  that, under certain conditions, infectious diseases may persist even when the basic reproduction number is less  than one. To provide further insights into control strategies, we analyze the structure of endemic equilibria in a  susceptible-infected-susceptible epidemic patch model in which the basic reproduction number is independent of  the dispersal rate of susceptible individuals. Our analysis demonstrates that variations in the dispersal rate of  susceptible individuals can significantly alter the structure of the endemic equilibrium set. In particular, we show  that, under some assumptions, the endemic equilibrium set forms a closed loop in the plane defined by the  susceptible dispersal rate and the total number of infected individuals at equilibrium. The biological implications  are significant. We establish that restricting the dispersal of susceptible individuals can substantially reduce disease  prevalence when the total population size is below a critical threshold. Conversely, when the population exceeds  this threshold, the disease is likely to persist. This work extends existing results by uncovering novel and intricate  structures within the endemic equilibrium set, thereby offering deeper theoretical insights into epidemic dynamics  and informing potential strategies for disease control.

Standing Wave Solutions to the Discrete Nonlinear Schrödinger Equation with Growing Potential

Maysoun Krnaf 

Morgan State University 

We consider the discrete nonlinear Schrödinger equation on the d-dimensional lattice with an infinitely growing potential, and  a sign changing general power-type nonlinearity, which is odd, continuously differentiable, and superliner near the origin, and  satisfies the Ambrosetti Rabinowitz condition. Our main analytic step is to prove that the associated Nemytskii derivative is compact from the energy space to its dual, which in turn yields the Palais–Smale condition for the corresponding variational  functional. We prove the existence of multiple exponentially decaying standing wave solutions via spectral splitting and linking  geometry. The multiplicities of solutions are expressed in terms of the frequency’s position relative to the spectrum and the sizes of the positive and zero regions of the coefficients in the nonlinearity. 

Compartmental model for analyzing sex trafficking dynamics 

Olawale Lawal 

Morgan State University 

This study develops a compartmental model for investigating the dynamics of sex trafficking. The population is divided into six  interacting compartments. The model describes the recruitment and spread of trafficking activities within the population.  Fundamental qualitative properties of the model, including positivity, boundedness, existence, and uniqueness of solutions, are  rigorously established to guarantee mathematical and epidemiological well-posedness. The sex trafficking-free and trafficking 

persistent (endemic) equilibria are derived and analyzed through the basic reproduction number, $R_0$, which determines  whether trafficking activities die out or persist in the population. Local and global stability analyses are carried out using the  Jacobian method, Lyapunov functions, and invariant region techniques. In addition, bifurcation analysis is performed to  investigate the qualitative changes in system dynamics near the critical threshold $R_0 = 1$, providing insights into the  conditions under which trafficking may persist even after substantial intervention efforts. Sensitivity analysis is further  conducted to identify the parameters with the greatest influence on trafficking transmission and persistence. The results  provide a quantitative foundation for designing effective intervention strategies such as awareness campaigns, rehabilitation programs, and strengthened law-enforcement policies for reducing trafficking prevalence and improving societal outcomes. 

Liouville PDE-based sliced-Wasserstein flow 

Pilhwa Lee 

Morgan State University 

The sliced Wasserstein flow (SWF), a nonparametric and implicit generative gradient flow, is transformed into a Liouville partial  differential equation (PDE)-based formalism. First, the stochastic diffusive term from the Fokker-Planck equation-based Monte  Carlo is reformulated as a Liouville PDE-based transport without the diffusive term, essentially reflecting the probability flow  ODE. The involved density estimation is handled by normalizing flows of neural ODE without an explicitly defined score  function. Next, the computation of the Wasserstein barycenter is approximated by the Liouville PDE-based SWF barycenter  with the prescription of Kantorovich potentials for the induced gradient flow to generate its samples. These two efforts show 

outperforming convergence in training and testing Liouville PDE-based SWF and SWF barycenters with reduced variance.  Applying the generative Liouville PDE-based SWF barycenter for fair regression demonstrates competent profiles in the  accuracy-fairness Pareto curves, with comparable and alternative choices against the standard SWF, and significant benefit in  improving fairness with scalability in comparison to the exact Wasserstein barycenter. 

A novel exploration of reasoning in Artificial Intelligence 

Dawn Lott 

Delaware State University, DE 

Artificial Intelligence (AI) focuses on enabling computer systems with the ability to perform tasks that previously only humans  primarily performed. As AI rapidly improves, there are still opportunities for incorporating other methodologies to enhance its  capabilities. One area is the continued look into different reasoning approaches. The goal would be to enhance Artificial  Intelligence by utilizing reasoning to achieve simple and complex tasks. By using additional different types of reasoning, AI will  continue to expand its ability to learn and adapt, solve problems, handle uncertainty, and interact with humans. In this paper,  we will explore state of the art for AI, different types of reasoning, what reasoning techniques are being used currently in AI,  and how other reasoning techniques can be leveraged in artificial intelligence. Because reasoning methods have proven to be  effective ways humans perform simple and complex tasks, this area of research still has potential opportunities for advancing the science. However, combining reasoning methods in AI may have associated challenges or limitations that may affect overall 

accuracy, reliability, or applicability in certain situations. In addition to investigating current AI and reasoning approaches we  plan to identify novel research directions and example use cases for uncovering new areas for future work.

Double-Inertial Adaptive CQ Methods for Split Variational Inequalities and Imaging Applications

Lovelyn Madu 

University of Louisiana at Lafayette 

This work introduces a double-inertial CQ-type projection and contraction method equipped with self-adaptive step sizes for  solving split variational inequality problems under a quasi-monotonicity framework in real Hilbert spaces. We establish strong  convergence of the proposed algorithm under mild and standard assumptions. Numerical experiments further demonstrate  that our method achieves superior performance compared to several existing algorithms in the literature, including additional 

comparisons on image processing and image restoration tasks. 

Gaussian Skewness Approximations for Dynamic Rate Multi-Server Queues 

William Massey 

Princeton University, NJ 

We introduce a three-dimensional dynamical system that estimates the transient mean, variance, and third cumulant moment  of a Markovian multi-server queueing system with non-homogeneous Poisson arrivals. This is a fundamental queueing model  for large scale service systems found in call centers and healthcare operations. These approximation methods surpass earlier  approaches by fitting the random number in the queue to a quadratic function of a Gaussian random variable. This is based on  papers published in QUESTA and is joint work with Jamol Pender of Cornell University. 

An Adaptive Iterative Approach to Stochastic Variational Inequalities with Applications

Temitope Mewomo 

Tarleton State University 

Stochastic variational inequality (SVI) problems arise naturally in the modeling of equilibrium phenomena under uncertainty,  with important applications in biological and health-related systems. We present an adaptive iterative scheme for  approximating solutions to SVIs, motivated by techniques from stochastic approximation, operator theory, and fixed-point  theory. The proposed framework is designed to enhance stability and performance in stochastic environments. Current work  focuses on establishing rigorous convergence properties, as well as exploring practical implementations and supporting  numerical experiments in application-driven settings. This is a joint work with Timilehin O. Alakoya (Queen's University Belfast,  United Kingdom). 

Enhancing Predictive Logistics: Using Optimized Machine Learning Models for Timely Delivery and Supply Chain  Efficiency 

Godgift Ndiwari 

Morgan State University 

In today’s world of globalized supply chains and rapid expansion of e-commerce, timely delivery is critical to developing  customer trust and happiness. This work aims to increase the prediction accuracy of the prompt delivery of consumer products, a key element for successful business operations, by improving machine learning approaches. To determine the activation  functions that best predict delivery timeliness, Extreme Learning Machine (ELM) and the Custom (ELM) architecture with  different activation functions were used. The results show that the choice of the activation function has a substantial impact.  While these variations may seem slight in large logistics networks, a tiny improvement can mean thousands of on-time  deliveries. We also emphasize model interpretability, i.e., the major drivers of delivery delays, in addition to prediction  precision. These observations offer actionable information for managers to take proactive measures to optimize operations and 

enhance the supply chain efficiency. The study demonstrates how improved machine learning models may be exploited to  combine technical innovation with real-world applications to enhance operational efficiency, reduce delays and ultimately  provide a more reliable consumer delivery experience. 

An Hymn to Periodicity 

Gaston N’Guerekata 

Morgan State University, Baltimore, MD, USA 

This talk aims to celebrate the world of periodicity through its diversity and applications. Periodicity is everywhere,  every day. Everybody is concerned with periodicity. We will show that there exist several types and generalizations  of periodicity motivated by real world phenomena. Applications to evolution equations, including the study of an  almost periodically forced pendulum, will be given.

Dynamics of solutions to a multiple-patch epidemic model with saturation incidence mechanism.

Cynthia Nnolum 

University of Nevada Las Vegas 

In this talk, I will present some of my research results on the dynamics of solutions to a multiple-patch epidemic  model with a saturation incidence mechanism. In particular, I identify explicit conditions on the model’s  parameters under which solutions to the initial value problem eventually stabilize. I also study the asymptotic  profiles of the endemic equilibrium solutions with respect to population dispersal rates. Numerical simulations  illustrating the theoretical results will also be presented. 

Reservoir–Immunity Thresholds, CTL Exhaustion, and Bifurcation Mechanisms of HIV Post-Treatment Control Magdaline Nchedo Nwankwor 

University of Louisiana at Lafayette 

Antiretroviral therapy suppresses HIV replication but does not eliminate the latent reservoir, so viral rebound  commonly occurs after treatment interruption. A subset of individuals, known as post-treatment controllers,  maintain low-level viremia without therapy, suggesting that remission depends on nonlinear interactions among  reservoir reactivation, immune clearance, and immune dysfunction. Building on the Conway--Perelson framework,  we develop a deterministic within-host HIV model that includes uninfected CD4^+ T cells, latently and productively  infected cells, free virus, effector CTLs, and CTL exhaustion pressure. The model incorporates infection, latent  activation, viral production, CTL-mediated killing, antigen-driven CTL expansion, and exhaustion-mediated  suppression through a Hill-type response. Analytically, we establish feasibility, derive the disease-free equilibrium  and basic reproduction number, and characterize chronic equilibria by reducing the steady-state system to a scalar  polynomial in the infected-cell level. We show that the polynomial threshold structure aligns with the biological  invasion criterion. Using CTL killing efficiency as a bifurcation parameter, we derive a critical immune-killing  threshold and identify conditions for forward or backward bifurcation. Numerical simulations confirm that stronger  CTL killing promotes control, while exhaustion thresholds shape bistability, hysteresis, and the transition between  rebound and remission. Latent reservoir burden at treatment interruption can determine whether the same  parameter regime leads to rebound, post-treatment-control-like suppression, or clearance-like control. Overall,  the results indicate that durable HIV remission depends on the balance among reservoir size, reactivation, CTL  strength, exhaustion dynamics, and the sharpness of immune suppression, providing a framework for evaluating  remission strategies that combine reservoir reduction, immune enhancement, and exhaustion mitigation. 

Mathematical Assessment of the Role of Vaccination on Chlamydia Control 

Jane Odeh 

Morgan State University 

This study presents a mathematical model (nonlinear differential equations) that incorporates Chlamydia, HIV, and  Chlamydia-HIV co-infection in humans. The model incorporates treatment for Chlamydia and HIV-infected  individuals and Chlamydia vaccination, which is under development. The model is rigorously analyzed, the basic  reproduction number is calculated, and the asymptotic stability property of the disease-free equilibrium is  established. Theoretical analyses revealed that the disease-free equilibrium is locally-asymptotically stable if the  basic reproduction number of the model is less than unity, the parameters related to the treatment, vaccination  rate and vaccination efficacy play a major role in bringing, and maintaining, the value of the basic reproduction  numbers below one (the epidemiological significance is that the disease can be eliminated if the basic  reproduction number of the model is less than one). The model is parametrized with data from Maryland.  Numerical simulation results show the importance of Chlamydia screening, access to treatment, vaccination  coverage, and efficacy. The research results showed that the prospects of effectively controlling Chlamydia  transmission are promising, with optimal vaccine coverage, vaccine efficacy, screening, and access to treatment

Quantitative Biology: Data Driven Modeling, Analysis and Computation in the Age of AI

Padmanabhan Seshaiyer 

George Mason University 

Quantitative biology is rapidly reshaping the study of living systems by integrating mathematical modeling,  computation, and data-driven analysis to better understand complex biological phenomena. Building on themes  developed in my recent book Quantitative Biology: Mathematical Modeling and Computation, this presentation  highlights a cohesive framework that combines analytical methods, statistical reasoning, and computational tools  to investigate biological and bio-inspired systems across multiple scales. The talk will examine a variety of modern  modeling strategies, including compartmental models, network-based approaches, and nonlinear dynamical  systems, with applications spanning epidemiology, biological interactions, and neural processes. Particular  attention will be given to the role of computation through numerical simulation, parameter estimation, and the  growing integration of machine learning with physics-informed modeling techniques for studying complex systems.  By connecting mathematical theory, computation, and real-world biological data, the presentation demonstrates  how quantitative approaches can contribute to addressing major societal challenges such as infectious disease  spread, health dynamics, and system resilience. More broadly, the talk emphasizes the expanding role of the  mathematical sciences in advancing modern biology while fostering interdisciplinary collaboration and innovation. 

Workshop: Computational Problem Solving with Neural Computing in the age of AI  

Padmanabhan Seshaiyer 

George Mason University 

In this workshop, we will introduce the foundations of computational problem solving for mathematical models  and datasets associated with describing real-world applications across science and engineering. Participants will  progressively engage with computational approaches across no-code, low-code, and high-code environments,  developing computational and algorithmic thinking through increasingly sophisticated tools and workflows. The  workshop begins with an introduction to computational thinking and exploratory data analysis, emphasizing how  computation can support modeling, simulation, and data-driven reasoning. We then provide an introduction to  machine learning fundamentals and demonstrate several state-of-the-art machine learning algorithms through  real-world applications. Building on these foundations, the workshop culminates with an introduction to neural  computing and artificial intelligence, highlighting how neural network architectures, scientific computing, and the  physical laws governing real-world systems can be integrated to support predictive modeling and data driven  decision making. 

Agent-based modeling of clustered cell migration in drosophila egg chambers 

Lara Scott 

UMBC 

Clustered cell migration is a process where a group of cells moves through biological tissue in response to a  stimulus, and it plays a crucial role in wound healing, tissue development, and cancer metastasis. We study  clustered cell migration using Drosophila egg chambers, which are widely used in biology due to their well understood genetic makeup and relatively short life cycle. In this system, a group of cells called border cells  migrates through the egg chamber during development, guided by chemical signals known as chemoattractants.  These chemical cues are challenging to study experimentally in vivo, so we create computational models that allow  us to study migration. My goal is to build a coupled model that better reflects how cells interact with both each  other and their chemical environment. To do this, I’m using Agent-Based Modeling (ABM)—a method where each  cell is made of boundary points that are being treated as independent “agents” following biologically motivated  rules. My current model simulates the physical forces that govern how cells stick to, repel, or push against each  other as well as how cells maintain their shape, allowing for realistic movement of a cell cluster through tissue. The  next step in my work is to couple this physical model with the chemical environment—so that cells respond  dynamically to the chemoattractant concentration around them, rather than moving along a chemical gradient  without well-known dependence on neighboring nurse cell adhesion. This would bring us closer to understanding  how collective decision-making emerges in biological systems.

Efficient Multilevel Methods for Material Properties Inversion in Heat Transfer Problems?

Alex Sheranko 

University of Maryland, Baltimore County 

Inverse problems for partial differential equations arise in many scientific and engineering applications and are often  formulated as optimal control problems. This presentation considers the inverse problem of recovering a spatially  varying thermal conductivity from the stationary heat equation. We develop a multigrid-based preconditioner to  accelerate the Gauss–Newton method used in the optimization process. The governing PDE is discretized using the finite  element method, while the multigrid framework exploits a hierarchy of coarser meshes to efficiently approximate the  resulting linear systems. Numerical results demonstrate that the proposed preconditioner has good approximation rates  and significantly reduces the number of Conjugate Gradient iterations required at each Gauss–Newton step. 

Stochastic Optimization and the Simultaneous Perturbation Method: Fundamentals and Selected Recent  Developments 

James Spall 

John Hopkins University, MD 

Stochastic optimization aims to minimize loss functions with only noisy function and/or gradient measurements.  Methods for stochastic optimization are used throughout virtually all areas of science and engineering. As a specific  example of a stochastic optimization algorithm, we will discuss the simultaneous perturbation stochastic approximation  (SPSA) algorithm for difficult multivariate optimization problems arising in stochastic systems and deep learning. The  essential feature of SPSA - which accounts for its power and popularity - is the underlying gradient approximation that  requires only two (noisy) objective function measurements regardless of the dimension of the optimization problem.  SPSA is particularly oriented to “zeroth-order” optimization, where gradients (stochastic or otherwise) are unavailable or  prohibitively expensive, including some problems in black-box learning and the fine-tuning of large neural network  models. Time permitting, we will discuss extensions of the basic SPSA method to: (i) stochastic versions of Newton’s  method, (ii) discrete or mixed variable problems, or (iii) multi-agent or feedback control problems involving dynamics.  This talk will focus on the basic ideas and motivation behind SPSA without dwelling on the mathematical details. 

Time-Fractional Damped Nerve Equation: Modeling and Q-HAM Solution 

Chidera Ugbonta 

Morgan State University 

Nerve signal propagation is essential for information transmission in biological systems and plays a significant role in  neuroscience, medicine, and bioengineering. Irregularities in nerve signal propagation are associated with several  neurological disorders, including Parkinson’s disease, epilepsy, multiple sclerosis, and neuropathic pain, thereby  motivating the need for accurate mathematical models that can capture complex neural behavior. In this work, we  propose a fractional-order damped nerve signal model that incorporates memory and hereditary effects to better  describe realistic neural behavior. Fractional calculus is used to capture nonlocal dynamics and anomalous signal  propagation in biological tissues. The Q-homotopy analysis method (Q-HAM) is employed to obtain analytical  approximations of the nonlinear fractional model. The results provide insight into the effects of damping and fractional order dynamics on nerve signal transmission and contribute to the mathematical understanding of neural disorders and  regenerative treatments such as spinal cord repair. 

A strongly convergent inertial-type algorithm for bilevel variational inequality problems

Odirachukwunma Ugwu 

Morgan State University 

Bilevel Variational Inequality Problems (BVIPs) arise in hierarchical decision-making processes where the upper-level decision  makers make strategic decisions while a lower-level system responds through an equilibrium modeled by a variational  inequality. These problems appear in several applications, including transportation networks, energy systems, image  processing, and resource management. Although projection-based methods have been widely studied for solving BVIPs, many  existing algorithms require multiple projections per iteration, leading to high computational cost. In this work, we propose an  efficient and strongly convergent algorithm for solving monotone BVIPs in Hilbert spaces. The proposed method incorporates  an inertial extrapolation step to accelerate convergence, an adaptive step-size strategy for improved stability and flexibility, and  a single projection onto a half-space to reduce computational complexity. Under mild assumptions, we establish strong  convergence of the proposed scheme. Numerical experiments demonstrate the effectiveness of the algorithm, showing faster  convergence and lower computational cost compared to several existing methods. Overall, this work provides an efficient and  practical framework for solving hierarchical optimization problems, with future extensions aimed at pseudomonotone settings,  large-scale applications, and data-driven optimization techniques.

Estimating Hidden Dynamics in Stochastic Reaction Networks from Partial Observations 

Mingkai Yu 

MEERI 

Stochastic reaction network models arise in intracellular reaction systems, as molecules interact through random  events. These systems can be modeled as continuous-time Markov chains whose state space represents the count  of the species involved. We consider the filtering problem with partial observations: estimating the conditional  probability distribution of a hidden state given exact partial state observations in two settings (continuous-time  and discrete-time observation). We propose weighted Monte Carlo particle filtering algorithms. We provide a  rigorous analysis as well as numerical examples to illustrate our method and compare it with alternatives. 

Sensitivity analysis of R0 in a mathematical model of Dermo disease in oyster population 

Najat Ziyadi 

Morgan State University 

Dermo disease that is caused by infection with Perkinsus marinus is an important factor of mortality for the oyster  population in affected areas. In this talk, a simple mathematical model of the spread of Dermo disease in oyster  population is presented as a system of ordinary differential equations. The basic reproduction number R0 is  computed using the next generation method. Sensitivity analysis will be used to illustrate the impact of model  parameters. 

Closure Approximations for the Erlang-A Queue 

Jhevon Smith 

Morgan State University 

The Erlang-A model is fundamental in queueing theory. It has many practical and theoretical applications, making  it widely studied by queueing theorists. It models the scenario of customers randomly joining a queue and waiting  for a random amount of time to obtain some service. After a while, each customer either receives the service or  abandons the queue due to impatience. Traditional methods for analyzing the Erlang-A queue include doing very  tedious analytical calculations and running Monte Carlo simulations for hundreds of thousands of iterations. We  provide an expedient alternative. We further develop a heuristic that first appeared in the 1998 paper "Strong  Approximations for Markovian Service Networks" by Mandelbaum, Massey, and Reiman, called the Closure  Approximation. This paper developed a “0-dimensional” analysis of the Erlang-A queue. Other papers in 2013 and  2018 pushed the analysis up to the “2-dimensional case.” We start from scratch and develop novel formulas and  techniques that allow us to build an algorithm that covers up to the “3-dimensional case” (and beyond, should one  possess the desire and pain tolerance). Ultimately, the tools we develop allow us to analyze the Erlang-A queue by  doing a single simulation run of a dynamical system. 

Mathematical Analysis and Simulation of Models for Cancer Therapy 

Xuming Xie 

Morgan State University 

Cancer remains one of the leading causes of death worldwide, prompting the exploration of innovative treatment  strategies. Among these, immunotherapy and oncolytic virotherapy have shown remarkable clinical success.  Mathematical models provide a systematic framework for studying complex biological interactions within tumors,  immune responses, and therapeutic interventions. Mathematical models of cancer immunotherapy commonly 

employ ordinary differential equations (ODEs) or partial differential equations (PDEs) to describe the interactions  between tumor cells, immune cells, and cytokines. In this talk, we will first review several deterministic and  stochastic ODE/PDE models for immunotherapy and virotherapy, and then we will analyze a stochastic free  boundary PDE model which includes noise terms (e.g., Wiener processes) to model uncertainty in tumor growth  rates, immune activation, and therapy effectiveness.