Thursdays, 4:30-5:30pm, Hunter East 920 or GC Room 6496 (currently hosted remotely; please email vrmartinez-at-hunter-dot-cuny-dot-edu for password)
HCAM was initiated by the late John Arthur Loustau, a former professor of the Mathematics and Statistics Department here at Hunter, with his then post-doc Emmanuel Asante-Asamani in 2018. John had an eclectic mix of mathematical interests, each of which he pursued with gusto and depth. In a career spanning nearly 50 years, he began his journey in Commutative Algebra, transitioned afterwards to Computer Science, then Numerical Analysis, and eventually into Mathematical Biology. John recounted once a vacation he took with his family to Reno long ago. His father, a hardware merchant who also did plumbing and electric work, had taken him to the School of Mining Engineering at University of Nevada and told John that he had once dreamt of enrolling there when he was younger. Nevertheless, as John fondly recalled,He was a fine applied mathematician.
In the spirit of John's open-mindedness and willingness to foster and maintain a diverse community of mathematics, HCAM hosts speakers across a wide range of disciplines, from both academia and industry. HCAM also showcases the work of rising graduates of the Applied Math MA program at Hunter and regularly hosts Applied Math MA alumni to share their post-graduate experiences with current students. If you are interested in giving a talk at this seminar, please contact me at vrmartinez-at-hunter-dot-cuny-dot-edu. Please note that this seminar is partially shared with the Nonlinear Analysis and PDEs seminar at the Graduate Center, so that some talks are hosted there instead.
Spring 2021 Schedule
February 25, 2021, Zoom (Recording)
Florian Mudekereza (Hunter Applied Math MA student, Adviser: Dana Sylvan)
Stochastic Statistical Inference in Games with Noisy Data
This talk concerns the development of a stochastic model for environments where players (producers) use statistical inference to form beliefs and make decisions. In the proposed setup, producers act as statisticians, each one obtains small noisy samples of the market supply, uses stochastic algorithms to improve their estimate of the market supply, and chooses to best-respond to this estimate. The proposed model relaxes the usual bounded rationality assumption by allowing producers to be aware of the noise and randomness in their data and stochastic estimates. As a result, this approach generates two key predictions which depend crucially on the sample size: asymptotically (i.e., in large samples), the stochastic estimates are shown to converge to the sampling equilibrium with statistical inference (SESI) market supply, Q_SESI; in small samples, there is a smaller market supply due to the low confidence level producers have in their stochastic estimates which leads to the proposed equilibrium concept called stochastic SESI (SSESI) Q_SSESI. Monte Carlo simulations illustrate the convergence of Q_SSESI to Q_SESI while their mean squared error is shown to be of order O(1/n). Potential improvements of this rate are explored.
March 4, 2021, Zoom (Recording)
Elizabeth Carlson (University of Nebraska-Lincoln)
Accurately Modeling Fluid Flow: Data Assimilation, Parameter Recovery, & Ocean Modeling
Scientists and mathematicians apply the continuum hypothesis to model fluid flow, i.e. the flow of substances like air or water. One of the challenges of the accurate simulation of turbulent flows is that initial data is often incomplete. Data assimilation circumvents this issue by continually incorporating the observed data into the model. In this talk, I will discuss my work using a new approach to data assimilation in order to both accurately model fluid flow and to identify certain physical properties of the fluid being modeled. I will also discuss implementation of this algorithm in large-scale climate models.
March 11, 2021, Zoom (Recording)
Owen Kunhardt (Hunter College Computer Science/Math BA student)
The Effects of Image Distribution and Task on Adversarial Robustness
Currently, few studies testing the theories of adversarial robustness take into consideration that when doing comparisons across each dataset, both the image distributions (e.g. digits vs objects) and classification task (to classify digits vs objects) are different. To unravel the causal factors of the inherent adversarial robustness of a model, we propose an unbiased metric to compare adversarial robustness and perform a series of experiments. In these experiments, we equalized several training hyperparameters on networks for the MNIST and CIFAR-10 datasets to determine whether the image distribution and task played a role in adversarial robustness. We find that networks trained for digit classification on MNIST digits are more adversarially robust than networks trained to do object classification on CIFAR-10 objects. In addition, to pin-point whether the contribution of adversarial robustness is mainly due to the image distribution or the task, we create a fusion image dataset that overlapped MNIST digits with CIFAR-10 objects such that image statistics were matched and train networks to perform a digit or object classification task. We find that models performing digit recognition on the fusion images were more robust than those performing object recognition, empirically verifying the role of the classification task in the adversarial robustness of a model, independent of the image distribution a network is trained on. Comparing the fusion model performances to their non-fusion counter-parts, we find that image distribution also plays a role.
March 18, 2021, Zoom (Recording)
Seckin Demirbas (University of British Columbia-Vancouver)
A Study on Certain Periodic Schrodinger Equations
This will be an informal introductory talk on the periodic cubic Schrodinger equation on 2-D irrational tori and the cubic fractional Schrodinger equation on the torus. We will discuss different tools and methods and how they ensure the local and/or global well-posedness of solutions to the equations at hand. We will also discuss the Sobolev norm growth for the global solutions of the periodic cubic Schrodinger equation on 2-D irrational tori.
March 25, 2021, Zoom (Recording)
Keisha Cook (Tulane University)
Single Particle Tracking with Applications to Lysosome Transport
Live cell imaging and single particle tracking techniques have become increasingly popular amongst the mathematical biology community. We study endocytosis, the cellular internalization and transport of bioparticles. This transport is carried out in membrane-bound vesicles through the use of motor proteins. Lysosomes, known for endocytosis, phagocytic destruction, and autophagy, move about the cell along microtubules. Single particle tracking methods utilize stochastic models to simulate intracellular transport and give rise to rigorous analysis of the resulting properties, specifically related to transitioning between inactive to active states. This confidence in the stochastic modeling of particle tracking is useful not only for particle-containing lysosomes, but also broad questions of cellular transport studied with single particle tracking.
April 8, 2021, Zoom
Greg Lyng (Optum Labs, UnitedHealth Group)
A year of COVID-19 models at OptumLabs
In this talk, we give a high-level (and selective) overview of the role that mathematical modeling has played in the OptumLabs response to the COVID-19 pandemic. This portfolio of models includes both national scale models of the spread of disease and local models aimed at informing surveillance testing and school/business reopening.
April 15, 2021, Zoom (Recording)
Pooja Rao (MSRI at UC Berkeley)
A novel strategy for unstructured search on IBM quantum processors
Grover's search is one of the most important quantum computing algorithms for unstructured search. However, when implemented on a real quantum device, the corresponding circuit depth increases prohibitively as the search domain gets bigger. A long circuit is undesirable as it leads to more quantum noise, which leads to poor results. In this talk, we introduce the basics of our algorithmic approach - based on quantum partial search - that reduces the depth of the quantum circuits significantly. With this approach, we have been able to design state-of-the-art circuits that not only show significant improvements over other existing results, but go a step further in being able to search through a bigger database.
April 22, 2021, Zoom (Recording)
Emmanuel Asante-Asamani (Clarkson University)
A mathematical model of weight change in humans
Weight management is of great concern to many Americans. People are either trying to lose weight or gain weight for health or aesthetic reasons. Often this endeavor is successful in the short term but fails in the long term. The question on the minds of most weight watchers is how to maintain the weight they have worked so hard to lose. Recently, a hormone secreted in the adipose tissues of mammals (leptin) was discovered and found to act on critical brain regions to control food intake and energy expenditure. Its primary goal is to maintain the body's fat stores. In this talk, I will present an extension of a mathematical model of weight dynamics to include the activity of leptin. The model explains why maintaining lost weight is so difficult and provides critical insight into how weight change can be maintained long term.
April 29, 2021 (Re-scheduled to Fall semester, September 23, 2021), Zoom
Aseel Farhat (Florida State University)
(TBA)
May 6, 2021, Zoom (Recording)
David Sondak (Harvard University)
Towards Bridging Machine Learning and Physics
A primary goal of science is to develop predictive models of physical phenomena. This is extraordinarily challenging, especially when multiscale physics is involved. Over the years, research has focused on trying to unlock the secrets of the governing equations as well as developing reduced models that capture the physical essence while being quick and easy to use. Recent research thrusts have started to bring machine learning algorithms to bear on classical physical problems such as fluid turbulence. This talk will provide an overview of an overview of neural networks and how they have been applied to physical problems. New results will be presented on an autoencoder neural network with a sparsity promoting latent space applied to canonical nonlinear partial differential equations.
May 13, 2021, Zoom
Cooper Boniece (Washington University in St. Louis)
A new stochastic process with multivariate covariance self-similarity
Stochastic processes that display aspects of scale invariance -- i.e., that possess features that remain unchanged under appropriate rescaling -- have been applied in a wide variety of disciplines ranging from hydrology to telecommunications to finance. However, by comparison to univariate models, far less attention has been paid to characteristically multivariate scale invariance models that display behavior that the limit theory arguably suggests is most natural. In this talk, I will discuss some mathematical background on scale invariant stochastic processes and some interesting aspects of a new family of processes called operator fractional Lévy motion. This is related to joint work with Gustavo Didier (Tulane University).
Current Seminar
Past Seminars