Jan Drgona

When and where:

Tuesday, October 28 | 1:15 PM - 2:30 PM
Session Title
Machine Learning for Optimization
Room
GWCC Building B Level 2 B201

talk info: submissions.mirasmart.com/InformsAnnua...
session info: submissions.mirasmart.com/InformsAnnua...

GitHub - pnnl/neuromancer: Pytorch-based framework for solving parametric constrained optimization problems, physics-informed system identification, and parametric model predictive control. Pytorch-based framework for solving parametric constrained optimization problems, physics-informed system identification, and parametric model predictive control. - GitHub - pnnl/neuromancer: Pyto...

I will be at the INFORMS this week.

I will be talking about a unifying perspective on Scientific Machine Learning for learning to optimize and learning to control, which is being materialized in the Neuromancer library:
github.com/pnnl/neuroma...

If you want to chat, send me a message :)

ROSEI is excited to unveil its first ever annual report, highlighting a year of rapid growth, groundbreaking research, and expanding impact! Explore the report and learn more about how the future of sustainable energy starts at JHU 🌎🔋💡 #HopkinsEnergy

Report: energyinstitute.jhu.edu/annual-repor...

I would like to thank Priya Donti for the invitation and all the faculty and students I met for wonderful discussions on computing and sustainability!

YouTube video by MIT Laboratory for Information and Decision Systems (MIT LIDS) 2025 LIDS Computing and Sustainability Seminar: Jan Drgona (Johns Hopkins University)

It was an absolute pleasure to give a talk at the Computing and Sustainability Seminar, hosted by MIT Laboratory for Information and Decision Systems (LIDS).

You can watch the recording here: www.youtube.com/watch?v=W9Lo...

This release was supported by the Ralph O’Connor Sustainable Energy Institute at Johns Hopkins University and by the EERE, BTO under the “Advancing Market-Ready Building Energy Management by Cost-Effective Differentiable Predictive Control” project.

Thanks to all contributors: Jan Boldocky, Aaron Tuor, Elad Michael

GitHub - pnnl/neuromancer: Pytorch-based framework for solving parametric constrained optimization problems, physics-informed system identification, and parametric model predictive control. Pytorch-based framework for solving parametric constrained optimization problems, physics-informed system identification, and parametric model predictive control. - GitHub - pnnl/neuromancer: Pyto...

See release updates here:
github.com/pnnl/neuroma...

GitHub - pnnl/neuromancer: Pytorch-based framework for solving parametric constrained optimization problems, physics-informed system identification, and parametric model predictive control. Pytorch-based framework for solving parametric constrained optimization problems, physics-informed system identification, and parametric model predictive control. - GitHub - pnnl/neuromancer: Pyto...

NeuroMANCER v1.5.6 is out!

We have four new examples:
1, Learning neural differential algebraic equations via the operator splitting method
2, Learning mixed-integer neural policies via DPC
3, Grid-responsive DPC for building energy systems
4, DPC with prediction preview horizon

This year’s workshop will feature more tutorials and hands-on coding examples — bridging theory and practice to bring physics-informed ML to life.

Organizers: Thomas Beckers, Truong Xuan Nghiem, Sandra Hirche, Rolf Findeisen, and Ján Drgoňa

Home Abstract

We’re excited to announce the 2nd Workshop on Physics-Informed Machine Learning at CDC 2025 in Rio de Janeiro on December 9th!

🔗 Workshop details: sites.google.com/view/2025cdc...
📝 Register here: cdc2025.ieeecss.org/registration
(Early registration ends soon!)

The Ralph O’Connor Sustainable Energy Institute at Johns Hopkins University also supported this research.

This research was supported by the U.S. Department of Energy through the Building Technologies Office under the Advancing Market-Ready Building Energy Management by Cost-Effective Differentiable Predictive Control project.

Learning Neural Differential Algebraic Equations via Operator Splitting Differential algebraic equations (DAEs) describe the temporal evolution of systems that obey both differential and algebraic constraints. Of particular interest are systems that contain implicit relat...

📜 Preprint: arxiv.org/abs/2403.12938
💻 Code: github.com/pnnl/NeuralD...

Thanks to all co-authors, James Koch, Madelyn Shapiro, Himanshu Sharma, and Draguna Vrabie, for their contributions. Special thanks to James for leading this work, full of follow-up research potential.

Our paper on "Learning Neural Differential Algebraic Equations via Operator Splitting" was accepted to the IEEE CDC.

If you don't have the time to read the full paper, check the paper summary with the Google Colab example at:
drgona.github.io/NeuralDAEs/

This release was supported by the Ralph O’Connor Sustainable Energy Institute at Johns Hopkins University.

🪲 Fixed Bugs:
Fixed plot issues in several examples
Fixed bug in CSTR system dynamics class in psl/nonautonomous.py
Fixed bug in computing constraints violations and objective function values in L2O examples

Kudos to Tyler Ingebrand for integrating the FE examples in Neuromancer!

Function Encoders: A Principled Approach to Transfer Learning in Hilbert Spaces A central challenge in transfer learning is designing algorithms that can quickly adapt and generalize to new tasks without retraining. Yet, the conditions of when and how algorithms can effectively t...

For more details on Function Encoders, check the papers by Tyler Ingebrand, Adam J. Thorpe, and Ufuk Topcu.
Function Encoders: A Principled Approach to Transfer Learning in Hilbert Spaces
arxiv.org/abs/2501.18373
Zero-Shot Transfer of Neural ODEs
arxiv.org/abs/2405.08954

GitHub - pnnl/neuromancer: Pytorch-based framework for solving parametric constrained optimization problems, physics-informed system identification, and parametric model predictive control. Pytorch-based framework for solving parametric constrained optimization problems, physics-informed system identification, and parametric model predictive control. - GitHub - pnnl/neuromancer: Pyto...

NeuroMANCER v1.5.4 is out!

github.com/pnnl/neuromancer

🆕 What's New in v1.5.4

💻 New Examples:
Function Encoders (FE) is an algorithm for learning neural network-based basis functions. Two new examples include the use of FE for function approximation and FE-Neural ODEs.

🚀 Less than two weeks to go until the 2025 American Control Conference in Denver, Colorado!

I’m excited for a packed schedule this year—with talks, a tutorial, a workshop, and most importantly, the chance to reconnect with friends and colleagues. 😊

sites.google.com/view/acc-phy...

GitHub - drgona/ML_and_control_buildings_energy Contribute to drgona/ML_and_control_buildings_energy development by creating an account on GitHub.

The spring semester is over, and my new course, "Introduction to Machine Learning and Control for Building Energy Systems," @jhu.edu is now complete.

The material, including lecture slides and code examples, is freely available on GitHub.

github.com/drgona/ML_an...

Home Abstract

Are you planning to attend the American Control Conference 2025 in Denver?

Consider joining our Workshop on Physics-Informed Machine Learning in Control: An Introduction, Opportunities, and Challenges

📅 Date: July 7, 2025
📍 Denver, Colorado
🔗 workshop details sites.google.com/view/acc-phy...

I am seeking a postdoctoral researcher to work on Scientific Machine Learning for Real-Time Decision Making.

📧 Interested candidates should send their CVs to my email: jdrgona1@jh.edu

Please see the job description below for more information.
www.linkedin.com/hiring/jobs/...

Huge thanks to all contributors to this release Rahul Birmiwal, Reilly Raab, Ali Reza Daneshvar Garmroodi.

GitHub - pnnl/neuromancer: Pytorch-based framework for solving parametric constrained optimization problems, physics-informed system identification, and parametric model predictive control. Pytorch-based framework for solving parametric constrained optimization problems, physics-informed system identification, and parametric model predictive control. - GitHub - pnnl/neuromancer: Pyto...

NeuroMANCER v1.5.3 is out!
github.com/pnnl/neuroma...

What is new?
🤖 NeuroMANCER-GPT Assistant
🐍 Python 3.11 Version Support
🏫 Building Control Comparison Example: Safe Reinforcement Learning vs Differential Predictive Control
💻 Improved Node Class

Our final #EWeek2025 post highlights Ján Drgoňa, who recently joined Hopkins as an associate professor in the Department of Civil and Systems Engineering! He was asked "How could your research make a future impact in the fight against climate change?" #HopkinsEnergy

Internship opportunity: Spatiotemporal Graph applications for Smart Buildings⚡⚡⚡ working closely with me, starting in early 2025. Considering applying or/and sharing this with your network please. Apply here: forms.gle/N3kwFxM3yEhS... LinkedIn ad here: www.linkedin.com/feed/update/...

ROSEI Researcher Q&A: Ján Drgoňa - Johns Hopkins - Ralph O’Connor Sustainable Energy Institute This article is part of a series featuring Q&As with Ralph O’Connor Sustainable Energy Institute (ROSEI)-affiliated researchers. Next up is Ján...

New ROSEI researcher Q&A with @drgona.bsky.social just went live! He discussed how a project as an undergraduate shaped his passion for energy-efficient building controls, and why joining the Hopkins energy community was a "no-brainer." #HopkinsEnergy

Story: energyinstitute.jhu.edu/rosei-resear...

Finite basis Kolmogorov-Arnold networks: domain decomposition for data-driven and physics-informed problems Kolmogorov-Arnold networks (KANs) have attracted attention recently as an alternative to multilayer perceptrons (MLPs) for scientific machine learning. However, KANs can be expensive to train, even fo...

FBKANs have been introduced in the paper:
Finite basis Kolmogorov-Arnold networks: domain decomposition for data-driven and physics-informed problems
authored by Amanda Howard, Bruno Jacob, Sarah Helfert Murphy, Alexander Heinlein, and Panos Stinis
arxiv.org/abs/2406.19662