Scientific

Wasserstein distances, or Optimal Transport methods more generally, offer a powerful non-parametric toolbox to conceptualise and quantify model uncertainty in diverse applications. Importantly, they work across the spectrum: from small uncertainty around a selected model (e.g., the empirical measure) to large uncertainty of considering all models consistent with the data. I will showcase this using examples from mathematical finance (pricing and hedging of options, optimal investment) and statistics (non-parametric estimators, regularised regression methods). I will illustrate the large uncertainty regime using Martingale OT problems. For the small uncertainty regime I will consider a generic stochastic optimization problem and its distributionally robust version using Wasserstein balls. I will derive explicit formulae for the first order correction to both the value function and the optimizer. Throughout, I will present both theoretical result, as well as comments on the available numerical methods.

The talk will be borrow from many joint works, including with Daniel Bartl, Samuel Drapeau, Stephan Eckstein, Gaoyue Guo, Tongseok Lim and Johannes Wiesel.

Large sets in Euclidean space should have large projections in most directions. Projection theorems in geometric measure theory make this intuition precise, by quantifying the words “large” and “most”.

How large can a planar set be if it contains a circle of every radius? This is the quintessential example of a curvilinear Kakeya problem, central to many areas of harmonic analysis and incidence geometry.

What do projections have to do with circles?

The talk will survey a few landmark results in these areas and point to a newly discovered connection between the two.

Monitoring marked individuals is a common strategy in studies of wild animals (referred to as mark-recapture or capture-recapture experiments) and hard to track human populations (referred to as multi-list methods or multiple-systems estimation). A standard assumption of these techniques is that individuals can be identified uniquely and without error, but this can be violated in many ways. In some cases, it may not be possible to identify individuals uniquely because of the study design or the choice of marks. Other times, errors may occur so that individuals are incorrectly identified. I will discuss work with my collaborators over the past 10 years developing methods to account for problems that arise when are only individuals are only partially identified. I will present theoretical aspects of this research, including an introduction to the latent multinomial model and algebraic statistics, and also describe applications to studies of species ranging from the golden mantella (an endangered frog endemic to Madagascar measuring only 20 mm) to the whale shark (the largest know species of fish, measuring up to 19m).

With the advancement in the high performance computing (HPC), it has become feasible to simulate various physical processes and phenomena. Such processes have applications ranging from energy & transportation sector to biological research. The process of liquid fuel injection and atomization forming fuel drops in aircraft engines is central to the formation of pollutants, therefore, it is crucial to study and control this process. The atomization is a physical process in which bulk liquid breaks up into small drops, further breaking up into even smaller drops finally leading to their evaporation. Quite often these drops are studied in an Eulerian fashion. Another approach to investigate the drops or deformable capsules is in a Lagrangian fashion. In this approach, each drop/capsule is tracked separately and is assumed to be either a rigid sphere or a deformable thin membrane. The latter has the direct application to the investigations of red blood cells (RBC) in biological systems. In fact, a RBC has a visco-hyperelastic thin membrane rendering it to be transported through capillary blood vessels of two times smaller its own size. By studying the dynamics of deformation of this membrane, it is possible to extract vital mechanical properties and develop a generalizable numerical model. This model has the potential to be employed to predict blocks in blood vessel the knowledge of which is helpful in improving the measurement of blood pressure. In this talk, I will be presenting two accurate, efficient, and robust numerical methods for simulating liquid fuel atomization process along with showcasing their engineering applications for subsonic & supersonic aircrafts. Furthermore, I will be giving a brief introduction to my current research work on the development of a numerical membrane model (NMM) for studying RBC deformation dynamics.

Convective storms are highly intermittent and intense, making their occurrence and strength difficult to predict. This is especially true for climate models, which have grid resolutions much coarser (e.g., 100 km) than the scale of a storm’s microphysical and dynamical processes (< 1 km). Physically-based parameterizations struggle to account for this scale mismatch, causing large model errors in rain and lightning. This talk will explore some avenues of using statistical techniques (such as generalized linear and log-Gaussian Cox process models) and machine learning methods (such as random forests and neural networks) that are trained by satellite observations of thunderstorms to see how well they can improve upon existing physical parameterizations in producing accurate rain and lightning characteristics given a set of large-scale environmental conditions.