Applications of Computational Math In Industry (CME 500)

Across the DOE, the wealth of data, robust automation, and stringent requirements for control, simulation, and data acquisition, make “Big Science” experiments — telescopes, particle accelerators, etc. — ideal targets for AI/ML. At the same time, the flavor of AI/ML techniques differ from those found in industry. In this talk, I will show some example AI projects at SLAC, including autonomous optimization, design, and anomaly detection in particle accelerators, and data analysis for single-particle imaging of biomolecules.

Daniel Ratner
Head of Machine Learning, SLAC National Accelerator Laboratory

Daniel Ratner completed his PhD in accelerator physics at Stanford in 2011, and then worked on various topics at SLAC including the commissioning of the first x-ray laser, particle accelerator-based EUV lithography sources, strong hadron cooling for colliders, and AI applications for the design and control of particle accelerators. Since 2019 he has been leading SLAC's lab-wide machine learning strategy, and is currently the ML department head. Before arriving at Stanford he worked at NYC’s MoMA using x-ray and optical methods to support art conservation. 

We developed a computational framework designed to predict how permafrost interacts with infrastructure, particularly in the face of climate change. Unlike existing models, our Arctic Coastal Erosion (ACE) framework incorporates advanced thermo-mechanical coupling to account for the complex relationship between heat flow, ice content, and mechanical behavior. This approach enables us to better capture key dynamics like subsidence and deformation caused by permafrost thaw.

We demonstrated the ACE framework’s capabilities by simulating settlement in a representative Arctic runway, showing how rising temperatures can compromise structural integrity. Given that Arctic infrastructure supports over four million people and 70% of permafrost-based structures, this work provides critical insights into failure risks and potential mitigation strategies. Ultimately, our framework aims to enhance resilience planning and address challenges in national security.

Alejandro Mota
Research Scientist, Sandia National Laboratories

Alejandro Mota is a Principal Member of the Technical Staff in the Mechanics of Materials Department at Sandia National Laboratories. He holds a PhD in Structural Engineering with a concentration in Theoretical and Applied Mechanics from Cornell University. His research focuses on computational methods for fracture, damage mechanics, and multi-physics simulations, with applications ranging from high-speed impact phenomena to Arctic coastal erosion. At Sandia, he develops advanced finite element methods, constitutive models, and multiscale techniques to improve simulations of material behavior under failure conditions.

Machine learning has led to innovative developments in many areas of earth sciences. Strikingly, in just a few years, data-driven weather forecasting has evolved from a proof-of-concept to fully operational medium-range AI forecasts served by leading meteorological agencies. Machine learning has the potential to further revolutionize climate science and weather forecasting across several domains including numerical model emulation, data-assimilation, long range forecasts, and high-resolution mesoscale weather forecasts. I will present some of our recent work on medium-range and storm-scale weather forecasting, ocean-coupled earth system modeling, and data assimilation while highlighting the challenges and open problems in the domain.

Jaideep Pathak
Senior Research Scientist, NVIDIA

Jaideep Pathak is a senior research scientist at NVIDIA focusing on large-scale machine learning for weather and climate modeling. He obtained his PhD in physics from the University of Maryland, College Park where he worked on fundamental problems in modeling chaotic dynamical systems using machine learning. Before joining NVIDIA, he worked on accelerating scientific models through deep learning and high-performance computing at the National Energy Research Scientific Computing Center (NERSC), a division of the Lawrence Berkeley National Laboratory (LBNL).

In this talk, we discuss the development of physically-constrained machine learning (ML) models that incorporate techniques from scientific computing for learning dynamical and physical systems with applications in epidemiology and fluid dynamics. We first study the lack of generalization of black-box deep learning models for ODEs with applications to COVID-19 forecasting and the need for incorporation of advanced numerical integration schemes. We then focus on learning a physical model that satisfies conservation laws which are ubiquitous in science and engineering problems ranging from heat transfer to fluid flow. Violation of these well-known physical laws can lead to nonphysical solutions. To address this issue, we propose a framework, which constrains a pre-trained black-box ML model to satisfy conservation by enforcing the integral form from finite volume methods. We provide a detailed analysis of our method on learning with the Generalized Porous Medium Equation (GPME), a widely-applicable parameterized family of PDEs that illustrates the qualitative properties of both easier and harder PDEs that is used in groundwater flow. Our model maintains probabilistic uncertainty quantification (UQ), and deals well with shocks and heteroscedasticities. As a result, it achieves superior predictive performance on downstream tasks, e.g., shock location detection. Lastly, we study how to hard-constrain Neural Operator solutions to PDEs to satisfy the physical constraint of boundary conditions on a wide range of problems including Burgers’ and the Navier-Stokes’ equations. Our model improves the accuracy at the boundary and better guides the learning process on the interior of the domain. In summary, we demonstrate that carefully and properly enforcing physical constraints using techniques from numerical analysis results in better model accuracy and generalization in scientific applications.

Danielle Maddix, MS '15 & PhD '18
Senior Applied Scientist in the Machine Learning Forecasting Group, AWS AI

Danielle Maddix Robinson is a Senior Applied Scientist in the Machine Learning Forecasting Group within AWS AI. She graduated with her PhD in Computational and Mathematical Engineering from the Institute of Computational and Mathematical Engineering (ICME) at Stanford University. She was advised by Professor Margot Gerritsen and developed robust numerical methods to remove spurious temporal oscillations in the degenerate Generalized Porous Medium Equation. She is passionate about the underlying numerical analysis, linear algebra and optimization methods behind numerical PDEs and applying these techniques to deep learning. She joined AWS in 2018 shortly after graduating, and has been working on developing statistical and deep learning models for time series forecasting. In this past year, she has been leading the research initiative on developing models for physics-constrained machine learning for scientific computing on the DeepEarth team. In particular, she has researched how to apply ideas from numerical methods, e.g., finite volume schemes, to improve the accuracy of black-box ML models for ODEs and PDEs with applications to epidemiology, aerodynamics, ocean and climate models.

This presentation will start with an overview of fair lending regulations and their influence on the use of machine learning in lending. It will then focus into statistical fairness metrics, emphasizing the practical challenges encountered during their implementation and interpretation. Finally, it will highlight strategies for enhancing the fairness of machine learning algorithms while preserving their accuracy. 

Enguerrand Horel, MS'16 & PhD '20
Sr. Machine Learning Manager, Upstart
Enguerrand Horel is a Senior Machine Learning Manager at Upstart. He received his PhD from the Institute for Computational and Mathematical Engineering at Stanford University. His research includes applying statistical learning techniques to financial problems such as credit risk prediction, and to develop principled ways to interpret machine learning estimators.

Max Zaraisky and Ben Bronselaer delve into the relationship between energy markets and weather. Max will begin with an exploration of the history of energy markets, highlighting the significant impact of weather over time. This historical perspective will provide a foundation for understanding the complexities and fluctuations within these markets. In the second part of the lecture, Ben will discuss the latest advancements in weather forecasting, emphasizing the role of modern tools and artificial intelligence. He will demonstrate how AI applications are transforming forecasting accuracy and their implications for energy market strategies. This session promises to offer valuable insights into the evolving landscape of energy markets and the cutting-edge technologies shaping their future.

Maxim Zaraisky  Partner and Commodities Portfolio Manager, Balyasny Asset Management (BAM)

Maxim Zaraisky is a Partner and Commodities Portfolio manager at Balyasny Asset Management (BAM). He joined BAM in 2021 from Squarepoint Capital, where he served as global head of gas and power. Before that, he was partner and European energy portfolio strategist at Cumulus Funds, and a European cross-commodity strategist at Merrill Lynch Commodities Europe. Zaraisky has an MBA from the Massachusetts Institute of Technology, a Bachelor of Arts in Economics from Durham University in the U.K., and completed the Stanford Executive Program at the Stanford Graduate School of Business.

Ben Bronselaer Analyst, Balyasny Asset Management (BAM)

Ben Bronselaer is an Analyst on Maxim Zaraisky’s team, in the London office. Ben joined BAM in December 2022 from Engelhart Commodity Trading Partners, where he was a Weather Analyst. Previously, he was with BP as a MetOcean, and Postdoctoral research scholar at NOAA GFDL, Princeton University and University of Arizona. He holds a DPhil from Oxford University and a Masters from Cambridge University.

Large Language Models (LLMs) have demonstrated remarkable capabilities across a wide range of tasks, but transitioning from monolithic predictors to versatile agents requires rethinking their architecture and interaction paradigms. In this talk, Hao Sheng will share insights into how post-training techniques, tool use, and system-level design can transform static models into dynamic agents. He will introduce the concept of Multi-Component Programs (MCPs)—a modular design pattern for orchestrating tool-augmented LLM behaviors—and discuss their role in enabling more reliable, compositional, and traceable reasoning. Drawing from both practical deployments and research prototypes, he will highlight lessons learned and open challenges in scaling these systems to meet real-world complexity.

Hao Sheng  MS'17 & PhD'21, Member of Technical Staff, OpenAI

Hao Sheng is a Member of Technical Staff at OpenAI, where he worked on RL, post-training research, and babysat ChatGPT. He earned his Ph.D. in Computer Engineering from Stanford University, working within the Stanford Machine Learning Group and the Stanford Computational Policy Lab under the guidance of Prof. Andrew Y. Ng and Prof. Sharad Goel. Hao has held roles at Apple’s Special Projects Group, TikTok, and Landing AI, where he led teams and built large-scale recommendation and generative AI systems. 

In lieu of in class lecture, student should plan on attending the ICME Research Symposium. This symposium provides a unique opportunity to explore how computational mathematics, data science, machine learning, scientific computing, and related fields are applied across various domains. 

Quantum computing has long promised revolutionary breakthroughs, but recent advances are rapidly shifting the conversation from potential to practice. In this talk, we will explore how quantum algorithms and quantum-inspired techniques are beginning to tackle real-world challenges across diverse sectors. From optimizing aircraft cargo loading in complex logistics networks, to managing power flows in modern energy grids, to accelerating large-scale linear algebra computations, and even fine-tuning large language models, quantum computing is quietly entering the commercial arena. In the final portion of the talk, I will briefly share my personal journey into quantum computing research in industry—highlighting the unexpected turns, the challenges of working at the cutting edge, and the opportunities available to graduate students curious about this rapidly evolving field.

Willie Aboumrad, Senior Quantum Applications Scientist

GEOS is an open-source framework for simulating tightly coupled thermal, hydrological, and mechanical processes in subsurface reservoirs, with a focus on geological carbon storage. Originally developed at Lawrence Livermore National Laboratory and redesigned under the DOE’s Exascale Computing Project, GEOS now benefits from contributions by Stanford University, TotalEnergies, Chevron, and others.

The framework supports modular physics packages, flexible coupling strategies, and performance portability across platforms, from laptops to GPU-based exascale systems. This talk highlights recent architectural innovations, current modeling capabilities, and results from challenging CO₂ storage simulations.

Bertrand Denel, Senior Research Scientist at TotalEnergies
Denel is currently based at Stanford University where he contributes to the FC Maelstrom project. He has 20 years of experience in geophysical research, specializing in numerical methods and high-performance computing for subsurface imaging. His previous work includes leading multiphysics and joint inversion initiatives to improve subsurface reservoir characterization, particularly for CO₂ sequestration monitoring. He received his Ph.D. in Applied Mathematics from the University of Pau, France, and completed a postdoctoral fellowship at the University of Houston.