The graduate student seminar is a biweekly seminar, alternating with the colloquium, in which students give talks about their current research. For the Fall 2024 semester, the graduate student seminars will be held on Wednesdays at 4:00 p.m in Crawford room 210.
For more information, or if you would like to give a talk, please contact Dr. Yakov Berchenko-Kogan at yberchenkokogan@doinghg.com.
Fall 2024
Initial Value Problems for Set Valued Differential Equations
November 20
Speaker
Uma Maheswara Rao Epuganti is a PhD student in Applied Mathematics at Florida Tech working with Dr. Gnana Bhaskar Tenali.
Abstract
Set-Valued Differential Equations (SVDEs) have connections to Fuzzy Differential Equations, multivalued differential equations, control problems, and many others. SVDEs are crucial for modeling systems with multiple possible outcomes, such as optimal control problems, classical mechanics, and economic models with uncertain trajectories. In classical mechanics, systems that experience impacts, such as a bouncing ball or multi-body collisions, can be modeled using SVDEs. For instance, after a collision, an object’s velocity may belong to a set of possible values due to variations in restitution laws or elastic properties. They also describe strategic decision-making in game theory, where players face a range of choices rather than a single optimal strategy.
After a brief overview of the definitions and properties of the Hukuhara, PS (Plotnikov-Skripnik), and BG (Bede-Gal) derivatives, we examine linear SVDEs involving each of these derivatives. Our main focus is on non-homogeneous linear SVDEs using the concepts of Hukuhara, PS, and BG derivatives. We obtain the solutions by transforming these differential equations into equivalent integral equations. It has been known that the Cauchy problem for SVDEs with PS and BG derivatives yields infinitely many solutions, among which two notable solutions—characterized by monotonically nondecreasing or decreasing diameter functions—are identified as basic solutions. We explore conditions under which these basic solutions exist, highlighting the role of the derivative type, coefficient matrix, initial set, and forcing set in determining solution behavior. To illustrate these results, we provide examples that emphasize the implications of our findings for SVDEs.
Integrating Fluid Dynamics into Systems Engineering: A Model-Based Approach to Hemodynamic Modeling
November 13
Speaker
Ceana Palacio is currently pursuing her M.S. in Systems Engineering at Florida Tech and is working with Dr. Luis Otero.
Abstract
Simulation modeling has become a valuable tool in healthcare for analyzing complex systems and optimizing patient care. In particular, fluid dynamics models are increasingly used to visualize hemodynamics and inform treatment decisions. One critical application is in the placement of left ventricular assist devices (LVADs), which provide mechanical circulatory support for patients with heart failure. However, current procedure guidelines for LVAD placement lack standardized guidelines on the optimal angle for pump positioning within the left ventricle, potentially impacting blood flow and device function. Fluid modeling can assist in determining the ideal placement by considering the cardiovascular geometry and optimizing blood flow, thus reducing the risk of misplacement during surgery. Although computational models can provide insights into optimal placement, their outputs are often challenging for clinicians to interpret. In this talk, we examine the application of fluid dynamics modeling through a systems engineering approach, utilizing open-source smoothed particle hydrodynamics (SPH) software to enhance the understanding and implementation of LVAD placement. By improving the interpretability of these models, this research aims to support more precise and effective surgical decision-making.
Critical p-Laplacian systems
October 16
Speaker
Abstract
p-Laplacian systems arise in the modeling of non-Newtonian fluids where the flow behavior depends on the rate of strain, leading to nonlinear flow characteristics and in image segmentation and denoising models that utilize the properties of p-norms to enhance image features. In the study of such problems, finding critical points of specific types of functionals is crucial and involves understanding the local behavior of saddle points of functionals defined in low-dimensional normed spaces and then extending these ideas to problems involving infinite-dimensional Banach spaces. This approach naturally leads to the application of algebraic topology to critical point theory, culminating in a beautiful overlap between topology and analysis. In this presentation, we will explore how these concepts can be applied to a certain type of critical p-Laplacian systems.
Higher-order blow-up finite elements
September 18
Speaker
Abstract
Finite element methods (FEMs) are numerical techniques for solving partial differential equations (PDEs) by reducing the space of solutions from an infinite-dimensional space to a finite-dimensional space. These methods are implemented in various fields, including structural analysis, fluid dynamics, heat transfer, electromagnetics, and biomechanics. Berchenko-Kogan and Gawlik recently introduced new finite element spaces, called blow-up finite elements. These spaces contain the classical finite element spaces but also allow certain kinds of singularities that make them conducive to simulating flows on curved surfaces, such as lava flows or the lubrication of mechanical parts. Berchenko-Kogan and Gawlik developed the lowest-order version of these spaces; in our work, we generalize these spaces to higher order. Higher-order finite element methods are coveted for improving the efficiency of implementations in terms of resources and accuracy. In this talk, we will go over the basics of finite element spaces, give an overview of the blow-up finite element spaces, and discuss the novel higher-order blow-up finite element spaces we propose.
References: http://yashabk.github.io/talks/blowupfeec-siam-handout.pdf
Spring 2024
Efficient Object Tracking with Applications to Visual Navigation
April 18
Speaker
Nehru Attzs is a graduate (PhD) student in Operations Research at Florida Tech working with Dr. Ryan White
Abstract
Object tracking is a fundamental problem in computer vision with numerous real-world applications, including visual navigation for autonomous systems. Limited by hardware computational and power budgets, efficient and robust object tracking methods are crucial for enabling these systems to perceive and reason about their environments. In this talk we will provide a brief overview of object tracking, looking at its application to visual navigation. We will also look at some ongoing research into methods for improving the current state of the art.
Nonlocal Boundary Value Problems for Linear Hyperbolic Systems of Second Order
April 11
Speaker
Afrah Almutairi is a graduate (PhD) student in Applied Mathematics at Florida Tech working with Dr. Tariel Kiguradze.
Abstract
MSE Graduate Student Seminar 2024-04-11 Abstract
Acoustic Machine Learning and Data Collection at the Edge
April 4
Speaker
Steven Wyatt is a graduate (PhD) student in Computer Engineering at Florida Tech working with Dr. Adrian Peter
Abstract
Good data is the essential first step towards a robust machine learning model yet it is often scarce in the field of environmental acoustics. This presentation outlines an iterative approach to deploying a machine learning model at the edge for data collection and model refinement. We will cover topics such as data collection, model training and finetuning, model conversion for deployment on edge devices and future plans for improvement.
Fall 2023
Existence of smooth solutions for the Landau equation with hard potentials
November 13
Speaker
Shelly Ann Taylor is a graduate (PhD) student in Applied Mathematics at Florida Tech working with Dr. Stanley Snelson
Abstract
The Landau equation was first documented in 1936, as an alternate version to the Boltzmann equation in the case of Coulomb interaction. It soon became (in combination with the Vlasov equation) the most important mathematical kinetic model in the theory of collisional plasma. Moreover, it is a nonlinear equation that combines transport in t and x, with non local diffusion in v. The aim of this presentation is to find a solution to the Landau equation that models the plasma particles. The solution f(t,x,v) will model the particle density of a plasma at time t > 0, location x in R3, and velocity v in R3.
Global existence with general initial data, poses a very challenging open problem. Therefore, the scope of this presentation primarily focuses on the nontrivial case of local existence. Particularly, the case of ``hard potentials,'' which is less understood than the other physical regimes, in terms of existence theory. Therefore, we prove the following: (i) precise upper bounds on the Landau coefficients in terms of weighted L∞ norms, (ii) pointwise lower bounds for solutions f for positive times, which is crucial for establishing ellipticity estimates for the diffusion matrix, and (iii) upper Gaussian bounds for solutions f of the Landau equation.
Time permitting, we will discuss plans for related future work.
Löewner, Legendre, and the Principal Pivot Transform - Oh My!
November 13
Speaker
Kenneth Beard is currently attending Louisiana State University pursuing a Ph.D. in Applied Mathematics. He obtained his Masters Degree in Applied Mathematics from Florida Tech.
Abstract
We present our recent results [1] on the (symmetric) principal pivot transform (PPT) improving on results in [2] which are motivated by the celebrated Löewner’s Theorem [3]. In particular, we give the minimum hypotheses for which the PPT is matrix monotonic with respect to the Löewner partial order on Hermitian matrices. This result turns out to have natural connections in the theory of Herglotz-Nevanlinna functions. In addition, we present a new variational principle for the PPT which is intricately connected with the Legendre-Fenchel transform from convex analysis. We then use this minimization principle to establish concavity of the PPT for kernel equivalent positive semi-definite matrices. Through- out we utilize the fundamental properties of Schur complements and the Moore-Penrose pseudoinverse under the Löewner ordering. This is joint work with Aaron Welters (Florida Institute of Technology).
[1] K. Beard and A. Welters, “Matrix monotonicity and concavity of the principal pivot transform,” arXiv, 2023. doi: 10.48550/arXiv.2302.04293.
[2] J. Pascoe and R. Tully-Doyle, “Monotonicity of the principal pivot transform,” Linear Algebra and its Applications, vol. 643, pp. 161–165, 2022. doi: 10.1016/j.laa.2022.02.016.
[3] B. Simon, Loewner’s Theorem on Monotone Matrix Functions. Springer Cham, 2019. doi: 10.1007/978-3-030-22422-6.
Diffusion Model based Super-resolution for Medical Images
October 30
Speaker
Mohammed Alsubaie is a graduate (Ph.D.) student in Operations Research at Florida Tech working under his advisor Dr. Xianqi Li
Abstract
High-resolution medical imaging is critical for precise diagnosis and optimal patient care. Yet, obtaining such images often necessitates prolonged scanning durations, leading to discomfort for patients and increasing the risk of scan failures due to patient movements. To address these challenges, super-resolution techniques were proposed and offer a promising avenue for enhancing the quality of medical images. Traditional super-resolution methods often fall short in preserving intricate details inherent to medical images. This talk introduces deep learning approaches, particularly the diffusion models, to enhance the resolution of medical images.
We begin by discussing the fundamental challenges in medical image super-resolution and the limitations of conventional methods. Then, we talk about the deep learning approaches for image super-resolution. In particular, we delve into the principles of diffusion models and shed light on why they are promising in this application. Experimental results will be showcased, comparing our diffusion model-based approach with other SOTA super-resolution methods. The outcomes reveal notable improvements in preserving fine-grained details and critical structures, making it an invaluable tool for clinicians.
Attendees will gain an in-depth understanding of diffusion models and how they can revolutionize the super-resolution of medical images, paving the way for enhanced diagnostic precision and patient outcomes.
Initial-Periodic Problems for Three-Dimensional Linear Hyperbolic Systems
October 16
Speaker
Najma Alarbi is a graduate (PhD) student in Applied Mathematics at Florida Tech working with Dr. Tariel Kiguradze.
Abstract
Bessmertnyĭ Realizations of Effective Tensors for Metamaterial Synthesis: Conjectures and Counterexamples
September 25
Speaker
Anthony is a graduate (Ph.D.) student in applied mathematics at Florida Tech working under his advisor Dr. Aaron Welters.
Abstract
We discuss the current progress towards my dissertation on the representability and analytic characterization of effective tensors for isotropic n-phase composites using a Hilbert space approach as described, for example, in [1] and [2]. The key properties of such tensors is that they are homogeneous degree-1 rational functions of n complex variables that belong to a special class known as Herglotz-Nevanlinna functions. My research is on proving the converse of functions with these properties, that is, the realizability problem for effective tensors (cf. [Ch. 7, 3]) which is important in metamaterial synthesis. Recently, a theorem was claimed by M. Bessmertnyĭ (in a preprint posted to arXiv [4]) to have been proven which would have essentially allowed us to do this. Unfortunately, we discovered that a certain lemma of his (attempting to rehabilitate a similar claim by T. Koga’s [5]), that he uses in the proof of this theorem, is false. We show this by constructing a counterexample (which also provides a counterexample, as we will show, to the result of T. Koga) that utilizes the stability properties of the basis generating polynomial to the Vámos matroid. These properties have been shown, over a decade, by several authors investigating matroids with the half-plane property and generalizations of the Lax Conjecture for hyperbolic polynomials. In this talk, we provide an overview of how these necessary properties for the counterexample were proved, such as being a homogeneous multi-affine stable polynomial and the existence of a certain real nonnegative polynomial that cannot be written as a polynomial sum-of-squares (SOS) [6] (which is interesting in its own right in light of Hilbert’s 17th problem [7] and the history associated with it). This talk is joint work with my advisor Dr. Aaron Welters.
References
- [1] G. W. Milton, The Theory of Composites. SIAM, Philadelphia, PA, 2023.
- [2] Y. Grabovsky, Composite Materials. Mathematical Theory and Exact Relations. IOP Publishing, 2016.
- [3] A. Stefan, Schur complement algebra and operations with applications in multivariate functions, realizability, and representations, M.S. Thesis, FIT, 2021.
- [4] M. F. Bessmertnyĭ, (2021). Representation of rational positive real functions of several variables by means of positive long resolvent. arXiv:2103.02541
- [5] T. Koga, (1968). Synthesis of finite passive n-ports with prescribed positive real matrices of several variables, in IEEE Transactions on Circuit Theory, vol. 15, no. 1, pp. 2-23. DOI:10.1109/TCT.1968.1082780
- [6] Wikipedia. 2008. “Polynomial SOS,” Wikimedia Foundation. Modified May 4, 2023. http://en.wikipedia.org/wiki/Polynomial_SOS
- [7] Wikipedia. 2005. “Hilbert’s seventeenth problem,” Wikimedia Foundation. Modified Sep. 2, 2023. http://en.wikipedia.org/wiki/Hilbert%27s_seventeenth_problem
Variational Discretization Methods for Curvature Flows on Riemannian Manifolds
September 11
Speakers
Zofia Adamska, Pedro Estrada Gallegos, and Ricardo Garcia
The speakers are students working on a research project in computational differential geometry with Dr. Yakov Berchenko-Kogan. Zofia Adamska and Ricardo Garcia are students from Caltech visiting under the Summer Undergraduate Research Fellowship (SURF) program and Pedro Estrada Gallegos is a student from Universidad Nacional Autόnoma de México.
Abstract
The study of geometric flows has several applications that span a wide variety of scientific disciplines. For example, these geometric evolution equations were used to model processes such as the annealing of metal sheets, evolution of reaction diffusion systems, or the behavior of cellular automata. Furthermore, geometric flows are applied to problems in computer vision and image processing. We present a variational method to compute curve shortening and curve straightening flows on arbitrary Riemannian manifolds. We apply our method to manifolds with different metrics and verify convergence to expected theoretical results. Furthermore, we introduce an algorithm that uses partitions of unity to evolve curves in manifolds covered by multiple charts, an example being noncontractible loops in tori. The methods developed in this project could have the potential to deepen our understanding of the geometry of non-trivial Riemannian manifolds, such as their geodesics and minimal submanifolds. These insights can open new avenues of investigation and inspire further developments in the field of discrete and numerical differential geometry.