Math-to-Industry Boot Camp XI

Overview

The Math-to-Industry Boot Camp is an intense six-week session designed to provide graduate students with training and experience that is valuable for employment outside of academia. The program is targeted at Ph.D. students in pure and applied mathematics. The boot camp consists of courses in the basics of programming, data analysis, and mathematical modeling. Students work in teams on projects and are provided with training in resume and interview preparation as well as teamwork.

The session features two group projects: a small-scale project designed to introduce the concept of solving open-ended problems and working in teams, and a "capstone project" that is posed by industrial scientists. Recent industrial sponsors included Cargill, Securian Financial and CH Robinson. Weekly seminars by speakers from many industry sectors provide the students with opportunities to learn about a variety of possible future careers.

Eligibility & Logistics

Applicants must be current graduate students in a Ph.D. program at a U.S. institution during the period of the boot camp. The program will take place online. Students will receive a $3,000 stipend.

Organizers

• Thomas Hoft, University of St. Thomas
• Daniel Spirn, University of Minnesota, Twin Cities

2026 Boot Camp Projects

Undersea Fiber Optic Sensing and Noise

Mentor: Derek Kane, SubCom

Beneath the ocean about 1.4 million kilometers of fiber optic cable carries 95% of all internet traffic. Beyond their value in securing the flow of information which connects all the world’s populations, these cables provide a largely untapped suite of sensors to track temperature, earthquakes, tsunamis, ships and marine wildlife. Light from a semiconductor laser transmitted down the fiber scatters from minute imperfections in the fiber’s material. Nominally, this purely random signal is fixed in time; however, mechanical stress from acoustic pressure waves or seismic events shifts the refractive index allowing an interferometer to detect the changes. By layering a temporal signal, such as a chirp, onto the laser light the mechanical changes can be isolated in time and space along the fiber optic cable.

We want to characterize the signal and the noise in this distributed acoustic sensing (DAS) system. The signal is the convolution of the laser’s interrogation signal convolved with a purely random return from the fiber, which in the absence of a mechanical disturbance, is fixed in time. What is the ideal temporal signal received, and how can a mechanical disturbance be detected? Once we describe the signal, we must analyze the noise. One source of noise is the Langevin fluctuations of the laser signal, a process modeled by a stochastic differential equation. Another noise-like source is the effect of the environmental disturbance along the cable on the signal. This is both what we want to monitor and a confounder of simple models of the signal. Beyond the questions posed by this problem, the techniques for analyzing the stochastic process have broad application in physical, financial and social systems.

Tax-Aware Portfolio Optimization

Mentor: Jing Wang, Thrivent

For an Exchanged Traded Fund (ETF), managing realized capital gains and loss is a very important part of portfolio management.  This project studies how to build an optimized rebalance portfolio that accounts for realized gains and losses with the goal of reducing immediate tax liability, using ideas inspired by tax-aware portfolio construction and convex optimization. In practice, this means balancing several competing objectives at once: maintaining portfolio alignment, limiting turnover and transaction costs, and making tax-efficient trading decisions.

Participants will start with a portfolio of stocks, represent holdings as tax lots, an alpha model and risk model and run monthly portfolio rebalance with and without consideration of tax liability, then compare tax-aware and tax-unaware trading approaches. They will explore how selling decisions depend on lot-level tax information such as cost basis and holding period, and how these details make the optimization problem more complex than a standard portfolio model. The project combines equity portfolio theory, convex optimization, and practical tradeoffs among tracking error, transaction costs, and taxes.

ABS Challenge System Analysis

Lee Przybylski, Washington Nationals

In 2026, Major League Baseball introduced a system where teams can challenge an umpire’s call of "Ball" or "Strike" using an Automated Ball Strike (ABS) system. This comes after a few seasons where the rule was implemented experimentally in AAA. The rules require that the challenge be made immediately after the call is made and only the batter, catcher, and pitcher may challenge. The challenge involves checking the horizontal and vertical coordinates of the ball as it passes a plane even with the middle depth of home plate. Each team is allowed to challenge twice, but if their challenge is upheld, they retain that opportunity to challenge again.

Going back to the introduction of pitch tracking data (beginning in 2016), it has been understood that catchers generate defensive value when they can receive a pitch in a way that makes the umpire more likely to call a pitch a strike instead of a ball. This skill is popularly referred to as framing. We want to study how the new challenge system will impact a catcher’s framing value.
The process should start by creating a statistical model that estimates the probability of a pitch being called a strike independent of the pitcher, catcher, batter, and umpire. Using this baseline, we can estimate the contributions of each party. To be more precise, suppose X equals 1 if a pitch, that is not swung at, is called a strike and 0 if it is called a ball. For a given player or umpire, we would like to know the change in E[X] after conditioning on that player or umpire’s involvement. This might be quantified in terms of log-odds first and then converted to an increase in strike probability. After that, we can convert the increase in strike probability to a run value using the RE288 matrix.

Identifying geological structures from multiple sources

Mentor: Melanie Jensen, SLB

How can we reconstruct the geometry of subsurface geological features from multiple, imperfect sources of information? Some measurements provide highly reliable but spatially sparse observations, while others provide broader coverage at the cost of uncertainty, ambiguity, missed detections, and false positives. The underlying geological structures themselves may exhibit complex geometries that cannot be adequately described by simple planar models. The goal of this project is to develop mathematical approaches for fusing heterogeneous observations, quantifying uncertainty, and inferring the most likely three-dimensional geological interpretation. This work sits at the intersection of applied mathematics, statistics, geometry, and machine learning, and has direct applications in subsurface imaging and characterization.

Components of this project include exploring various techniques to identify which measurements correspond to the same 3D-geological structure, and if we can develop artificial intelligence to select the technique that should be used. This will include feature selection, model selection, and model evaluation. We will use synthetic data with varying levels of noise to capture realistic measurement data.