Matt Baughman
Postdoctoral Computational Scientist // Princeton University & PPPL
Hi there! I'm a Postdoctoral Computational Scientist at Princeton, working within the DOE's Princeton Plasma Physics Laboratory with Shantenu Jha. Most of my current work is part of the DOE's Genesis Mission, building multi-agent systems that run scientific campaigns end to end: proposing experiments, executing simulations on leadership-class machines, and evaluating what comes back. I also work on the skill and context mechanisms those systems need to stay tractable as they grow. Before Princeton, I completed my Ph.D. in Computer Science with Globus Labs at the University of Chicago in August 2025, advised by Ian Foster and Kyle Chard. My dissertation was on cost-aware placement of work across heterogeneous, federated systems; I now work that same problem one layer up, with agents as the thing being placed. I completed my Bachelor's in Computer Science and Philosophy at Minerva University, and have worked at Argonne National Laboratory, Opinary, and Google.
science RESEARCH link
chevron_right
Distributed Agentic Systems: We are using AI agents to dynamically scale science, driving simulation, experimental, and discovery campaigns that decide for themselves what to run next and grow to meet the problem rather than the other way around.
chevron_right
AI for Science: We are putting foundation models and learned surrogates for fusion energy to work inside real optimization loops, coupling them to the simulation codes they stand in for and to the facilities those codes run on.
chevron_right
Cost-aware Computing: We are building scalable frameworks for highly distributed science workflows spanning dozens of computing sites and scaling to tens of thousands of compute nodes using novel cost-aware task placement algorithms and federated function-as-a-service infrastructure.
engineering PROJECTS link
chevron_right
Tritium Splash: An agentic control loop for tritium-breeding blanket design, searching for configurations that sustain their own fuel supply. A campaign layer proposes and evaluates experiments while an inner orchestrator executes a declarative graph over neutronics, chemistry, materials, and quantum agents, each backed by a surrogate or the real simulation code depending on what a given facility can run.
[Code]
[Video]
chevron_right
IGNOS: An evolutionary and Bayesian outer loop over plasma profile parameters for core-edge integration, running one candidate per GPU against diffusion-model imputation and simulation surrogates, with every campaign and candidate tracked as nested experiment runs.
chevron_right
Genesis Skills: A portable agent-skill catalog for the DOE Genesis Mission, plus a dependency-free search engine that loads skills progressively instead of flat-loading the catalog into context, cutting the context cost of the catalog by roughly 19x. Skills must run from any agent harness, so nothing harness-specific is admitted.
chevron_right
Genesis Agent Skills Onboarding Demo: A single prompt plus a few agent skills that get any coding agent to build a working multi-agent optimization campaign: competing search strategies measured against a hidden response surface, with a skeptical aggregator that re-measures each claimed result before trusting it. Verified by clean-room rebuilds from the prompt alone.
[Code]
star SELECTED PUBLICATIONS link
Ordered by most recent.
chevron_right
| Adaptive Task Management: Automating Workload Distribution Across Computing Sites [Jul 2026] |
| Matt Baughman |
| In submission |
| TLDR | BibTeX |
|
TLDR: Compute is the currency of modern science, but there is no unified way to spend it across the many systems that make up the computing continuum. Adaptive Task Management (ATM) is a multi-system task manager that maps tasks onto resources spanning edge devices, cloud, and batch-scheduled HPC, combining a cost-aware placement algorithm with monitoring and management designed for both large batches of independent tasks and DAG-based workflows.
|
@unpublished{baughman2026atm,
title = {Adaptive Task Management: Automating Workload Distribution Across Computing Sites},
author = {Matt Baughman},
note = {In submission},
year = {2026}
}
|
chevron_right
| Limitless, With an Asterisk: Agent Scaling and Skills for the Modern Era of Computational Science [Jun 2026] |
| Matt Baughman |
| In submission |
| TLDR | BibTeX |
|
TLDR: Most scientific agents today are hand-crafted demonstrations: orchestration is bespoke, capabilities are re-derived project by project, and one team's execution experience never compounds into another's. We argue that two paradigms convert demonstrations into reliable capability: hierarchy-agnostic agent interfaces, which let agents nest recursively so scale is no longer bounded by architecture, and skills, portable capability packages distilled from execution traces that let what agents learn travel across agents, projects, and institutions. We close on the asterisk on limitless scale: the verification, autonomy, cost, evaluation, and institutional-memory constraints that architecture alone does not solve.
|
@unpublished{baughman2026limitless,
title = {Limitless, With an Asterisk: Agent Scaling and Skills for the Modern Era of Computational Science},
author = {Matt Baughman},
note = {In submission},
year = {2026}
}
|
chevron_right
| Flight: A FaaS-based framework for complex and Hierarchical Federated Learning [Jan 2026] |
| Nathaniel Hudson, Valérie Hayot-Sasson, Yadu Nand Babuji, Matt Baughman, J. Gregory Pauloski, Ryan Chard, Ian T. Foster, Kyle Chard |
| Future Generation Computer Systems |
| TLDR | Publication | BibTeX |
|
TLDR: Flight is a federated learning framework built on function-as-a-service that supports complex hierarchical and asynchronous topologies, decoupling FL logic from infrastructure so the same workflow runs from edge devices to HPC.
|
@article{hudson2026flight,
title = {Flight: {A} FaaS-based framework for complex and Hierarchical Federated Learning},
author = {Nathaniel Hudson and Val{\'{e}}rie Hayot{-}Sasson and Yadu Nand Babuji and Matt Baughman and J. Gregory Pauloski and Ryan Chard and Ian T. Foster and Kyle Chard},
doi = {10.1016/J.FUTURE.2025.107998},
journal = {Future Gener. Comput. Syst.},
pages = {107998},
url = {https://doi.org/10.1016/j.future.2025.107998},
volume = {174},
year = {2026}
}
|
chevron_right
| Trillion Parameter AI Serving Infrastructure for Scientific Discovery: A Survey and Vision [Dec 2023] |
| Nathaniel Hudson, J. Gregory Pauloski, Matt Baughman, Alok Kamatar, Mansi Sakarvadia, Logan T. Ward, Ryan Chard, André Bauer, Maksim Levental, Wenyi Wang, Will Engler, Owen Price Skelly, Ben Blaiszik, Rick Stevens, Kyle Chard, Ian T. Foster |
| BDCAT 2023 |
| TLDR | Publication | BibTeX |
|
TLDR: We survey the infrastructure required to serve trillion-parameter AI models for science and lay out a vision for integrating such models with HPC facilities.
|
@inproceedings{hudson2023trillion,
title = {Trillion Parameter {AI} Serving Infrastructure for Scientific Discovery: {A} Survey and Vision},
author = {Nathaniel Hudson and J. Gregory Pauloski and Matt Baughman and Alok Kamatar and Mansi Sakarvadia and Logan T. Ward and Ryan Chard and Andr{\'{e}} Bauer and Maksim Levental and Wenyi Wang and Will Engler and Owen Price Skelly and Ben Blaiszik and Rick Stevens and Kyle Chard and Ian T. Foster},
booktitle = {Proceedings of the {IEEE/ACM} 10th International Conference on Big Data Computing, Applications and Technologies, {BDCAT} 2023, Taormina (Messina), Italy, December 4-7, 2023},
doi = {10.1145/3632366.3632396},
pages = {15:1--15:10},
publisher = {{ACM}},
url = {https://doi.org/10.1145/3632366.3632396},
year = {2023}
}
|
chevron_right
| Accelerating Communications in Federated Applications with Transparent Object Proxies [Nov 2023] |
| J. Gregory Pauloski, Valérie Hayot-Sasson, Logan T. Ward, Nathaniel Hudson, Charlie Sabino, Matt Baughman, Kyle Chard, Ian T. Foster |
| SC 2023 |
| TLDR | Publication | BibTeX |
|
TLDR: ProxyStore introduces transparent object proxies that decouple bulk data movement from control flow in federated applications, accelerating communications across distributed executors.
|
@inproceedings{pauloski2023proxystore,
title = {Accelerating Communications in Federated Applications with Transparent Object Proxies},
author = {J. Gregory Pauloski and Val{\'{e}}rie Hayot{-}Sasson and Logan T. Ward and Nathaniel Hudson and Charlie Sabino and Matt Baughman and Kyle Chard and Ian T. Foster},
booktitle = {Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis, {SC} 2023, Denver, CO, USA, November 12-17, 2023},
doi = {10.1145/3581784.3607047},
pages = {59:1--59:15},
publisher = {{ACM}},
url = {https://doi.org/10.1145/3581784.3607047},
year = {2023}
}
|
chevron_right
| Balancing Federated Learning Trade-Offs for Heterogeneous Environments [Mar 2023] |
| Matt Baughman, Nathaniel Hudson, Ian T. Foster, Kyle Chard |
| PerCom Workshops 2023 |
| TLDR | PDF | Publication | BibTeX |
|
TLDR: We characterize the accuracy, cost, and time trade-offs inherent to federated learning across heterogeneous devices and show how informed configuration improves outcomes.
|
@inproceedings{baughman2023balancing,
title = {Balancing Federated Learning Trade-Offs for Heterogeneous Environments},
author = {Matt Baughman and Nathaniel Hudson and Ian T. Foster and Kyle Chard},
booktitle = {{IEEE} International Conference on Pervasive Computing and Communications Workshops and other Affiliated Events, PerCom Workshops 2023, Atlanta, GA, USA, March 13-17, 2023},
doi = {10.1109/PERCOMWORKSHOPS56833.2023.10150228},
pages = {404--407},
publisher = {{IEEE}},
url = {https://doi.org/10.1109/PerComWorkshops56833.2023.10150228},
year = {2023}
}
|
chevron_right
| Rural AI: Serverless-Powered Federated Learning for Remote Applications [Mar 2023] |
| Panos Patros, Melanie Ooi, Victoria Huang, Michael Mayo, Chris Anderson, Stephen Burroughs, Matt Baughman, Osama Almurshed, Omer F. Rana, Ryan Chard, Kyle Chard, Ian T. Foster |
| IEEE Internet Computing |
| TLDR | Publication | BibTeX |
|
TLDR: We make the case for serverless-powered federated learning to bring AI to remote and rural applications, outlining an architecture and open challenges for intermittently connected environments.
|
@article{patros2023ruralai,
title = {Rural {AI:} Serverless-Powered Federated Learning for Remote Applications},
author = {Panos Patros and Melanie Ooi and Victoria Huang and Michael Mayo and Chris Anderson and Stephen Burroughs and Matt Baughman and Osama Almurshed and Omer F. Rana and Ryan Chard and Kyle Chard and Ian T. Foster},
doi = {10.1109/MIC.2022.3202764},
journal = {{IEEE} Internet Comput.},
number = {2},
pages = {28--34},
url = {https://doi.org/10.1109/MIC.2022.3202764},
volume = {27},
year = {2023}
}
|
chevron_right
| FLoX: Federated Learning with FaaS at the Edge [Oct 2022] |
| Nikita Kotsehub, Matt Baughman, Ryan Chard, Nathaniel Hudson, Panos Patros, Omer F. Rana, Ian T. Foster, Kyle Chard |
| eScience 2022 |
| TLDR | Publication | BibTeX |
|
TLDR: FLoX runs federated learning on serverless (FaaS) infrastructure at the edge, decoupling model training and inference from deployment so FL can run on heterogeneous real-world devices.
|
@inproceedings{kotsehub2022flox,
title = {FLoX: Federated Learning with FaaS at the Edge},
author = {Nikita Kotsehub and Matt Baughman and Ryan Chard and Nathaniel Hudson and Panos Patros and Omer F. Rana and Ian T. Foster and Kyle Chard},
booktitle = {18th {IEEE} International Conference on e-Science, e-Science 2022, Salt Lake City, UT, USA, October 11-14, 2022},
doi = {10.1109/ESCIENCE55777.2022.00016},
pages = {11--20},
publisher = {{IEEE}},
url = {https://doi.org/10.1109/eScience55777.2022.00016},
year = {2022}
}
|
book THESES link
chevron_right
| Corralling the Computing Continuum: Enabling Multi-System Workflows with Serverless Computing [Aug 2025] link |
| Abstract | Committee | PDF | Slides | Doctoral Dissertation |
|
ABSTRACT: The computing continuum describes the convergence of global compute infrastructure as network bandwidths increase. To mobilize that infrastructure, we need to create a system that ties these diverse resources together—we need to corral the computing continuum. This effort began with task-wise solutions, addressing different components of task placement—profiling, predicting, and provisioning. These individual solutions enabled an early system that took into account compute costs, workload execution profiles, and the ability to move compute between systems. We combined and extended these works into a more robust task scheduling system called DELTA and its successor DELTA+. These systems incorporated notions of task execution time, data transfer costs, and machine performance but could not be used on batch scheduled systems or in multi-node environments. While compute is the currency of the future, there is no unified way to access that currency. To fill this gap, we introduce Adaptive Task Management (ATM)—a framework that acts as a multi-system task manager, mapping tasks to the many resources that comprise the continuum. ATM is designed on top of the Globus Compute framework, using existing infrastructure from edge devices to batch-scheduled HPC systems. ATM includes a novel placement algorithm and novel monitoring and task management systems designed to accommodate both large batches of tasks as well as more complex DAG-based workflows. To ground and evaluate the development of these frameworks, we explore the application of cost-aware principles in federated learning, material design and protein docking science applications, and in the performance optimization of serverless computing benchmarks.
|
|
Committee: Kyle Chard, Ian Foster, and Omer Rana (Cardiff University)
|
chevron_right
| Profiling, Predicting, and Provisioning: Enabling Cost-Aware Computation for the Cloud and Modern Heterogeneous Environments [Aug 2021] link |
| Abstract | Committee | PDF | Slides | Master's Thesis |
|
ABSTRACT: The growing prevalence of cloud resources and specialized hardware in the form of GPUs, ASICs, and IoT devices requires increasingly efficient and intentional use of these resources. Moreover, the complexity of choice presented by these diverse resources creates an optimization problem largely intractable to manual control. Therefore, modern computation in heterogeneous environments must be executed in a cost-aware, automated fashion. This control system can be decomposed into three discrete tasks: profiling, prediction, and provisioning. We profile the execution characteristics of a range of workloads on a range of hardware. Given those characteristics, we optimize our choice of resources for workload deployment based on predicted cost. Finally, we seamlessly provision any necessary resources and deploy the workload given the optimized choice of resource. In this thesis, we integrate several projects spanning the profiling, predicting, and provisioning cycle towards a unified system for the cost-aware distribution of workloads in dynamic, heterogeneous computing environments. Specifically, we develop a modular profiling system that characterizes the execution performance of scientific workflows deployed on cloud resources, employ statistical analyses and machine learning to predict the cost of using preemptible cloud resources, examine the role of computational tradeoffs in various workloads, and build on existing Function-as-a-Service (FaaS) frameworks to demonstrate a novel, cost-aware function distribution system. Through this work, we show the significant cost and time reductions for scientific workload execution, while enabling function-based distributed computing in a cost-aware heterogeneous environment.
|
|
Committee: Kyle Chard, Ian Foster, and Hank Hoffmann
|
explore SERVICE & INTERESTS link
chevron_right
Minerva University: I volunteer with my undergraduate alma mater as a guest instructor and on senior thesis committees, serve on the alumni council, and mentor students.
chevron_right
DSI Summer Lab: From 2021 to 2025 I coordinated the University of Chicago Data Science Institute's Summer Lab part-time, growing it from a dozen students recruited mostly by personal referral into one of the most selective data science research programs for high schoolers and undergraduates in the country.
chevron_right
Aviation: I had my pilot's license before my driver's license, have worked on building three different planes, and still make it to EAA every year.