About
Howard builds and deploys algorithms that combine physics-based modeling with data to maximize the performance of complex, constrained systems. He has worked with teams in medical imaging, e-commerce, physics-based animation, and trading crypto on blockchains. Through Typal Academy, he shares insights in the field of optimization and its interface with deep learning.
| Credential | Detail | Year |
|---|---|---|
| Ph.D. Mathematics | Thesis: Learning to Optimize with Guarantees | 2021 |
| M.A. Mathematics | Qualifying Exams: Numerical Analysis, Applied Differential Equations | 2018 |
| Credential | Detail | Year |
|---|---|---|
| B.S. Computer Science, Mathematics, Physics | GPA 3.97 / 4.00 | 2016 |
Howard's research originated in convex feasibility problems and iterative projection methods. In graduate school, this expanded to first-order optimization algorithms (e.g. operator splitting) and using machine learning to speed up algorithms and/or modify optimization problems to best utilize knowledge hidden in historical data (e.g. for inverse problems).
First-order and operator-splitting methods for large-scale convex optimization, with a focus on convergence guarantees, robustness and practical implementations.
Leveraging available data to solve optimization-based problems more effectively. This includes using deep learning to
This general area of research has many names: physics-informed machine learning, scientific machine learning, learning-to-optimize, predict-then-optimize and decision-focused learning.
Industrial Problems Seminar, IMA at University of Minnesota
Nov 21, 2025
Optimization problems and big data arise all over in industry. This talk will walk through examples of these problems and highlight how optimization solvers can be informed by data (e.g. decision-focused learning). Discussion will then transition to lessons learned from research positions in various fields.
AI Meets Optimization, ICCOPT Session at University of Southern California
Jul 22, 2025
Many real-world problems can be framed as optimization problems, for which well-established algorithms exist. However, these problems often involve key parameters that are not directly observed. Instead, we typically have access to data that is correlated with these parameters, though the relationships are complex and difficult to describe explicitly. This challenge motivates the integration of machine learning with optimization: using machine learning to predict the hidden parameters and optimization to solve the resultant problem. This integration is known as decision-focused learning. In this talk, I will introduce decision-focused learning, with a particular focus on differentiating through solutions to optimization problems and recent advances in effectively scaling these computations.
ACMD Seminar, National Institute of Standards and Technology
Mar 7, 2023
Flexible, human-interpretable machine learning models are gaining interest as applications increasingly require explainable artificial intelligence. This talk overviews recent developments in the "learn to optimize" (L2O) methodology wherein model inferences are defined to be solutions to parameterized optimization problems. The idea is to have domain experts create intuitive optimization models that include both analytic and parameterized terms. This fusion merges data-driven modeling with strong analytic guarantees (e.g. inferences satisfying linear systems of constraints). We will cover the key tools needed to design and implement L2O models along with numerical examples.
Someone that can take vague goals and constraints and own the process for developing algorithmic solutions in code that engineers can use. They also provide documentation and reports so people know how to use the math safely and interpret results, e.g. guides for engineers and clear summaries for leaders.
Howard writes short posts, each illustrating concepts in his area of work.
Howard hosts the podcast "Numerical Optimization" where he interviews mathematicians in various optimization specialties.
