About the Workshop
Machine learning is transforming engineering and physical sciences, enabling data-driven solutions across disciplines in estimation, forecasting, sensing, and control tasks. This workshop brings together domain experts, practitioners, and ML/AI developers to further innovation at the intersection of physics-informed AI and dynamic physical systems in the service of energy efficiency.
Hosted by the AI Institute in Dynamic Systems, AI-DEEDS aims to establish a community-driven framework for evaluating algorithms that address the complexities of dynamic systems in engineering and science.
Building off the success of AI-DEEDS 2025, this year we hosted 2 challenges on the CTF platform, where individuals and teams submitted and benchmarked their algorithms. The workshop took place on June 22, 2026 at ACM e-Energy — see the challenge results.
Workshop Goals
- Build a taxonomy of architectures for modeling dynamic systems
- Share datasets across disciplines focused on improving Energy Efficiency
- Foster a vibrant research community focused on next-generation ML tools for energy efficiency
Key Dates
CTF Challenges
Both challenges ran on our Common Task Framework and have now concluded — see the results below.
Smart Buildings Challenge
Design HVAC control agents using Google's open-source simulator. Minimize carbon emissions and operating costs while maintaining comfortable building conditions.
Buildings account for ~40% of US carbon emissions. Your solution could have massive environmental impact.
Learn MoreChaotic Systems Challenge
Advance research on modeling complex dynamical systems with applications across energy and beyond.
Learn MoreIndustry Partners
Providing the simulator for the interactive CTF challenge competition
Hosting the challenges with technical support and prize fund
Challenge Results
Both challenges have concluded. Congratulations to the winners, and thank you to everyone who competed and submitted papers!
🌀 Chaotic Systems Challenge
Michael Nawar
Horizon-aware Assimilation and Invariant-measure Sampling for Lorenz-63 Trajectories
More Challenge Papers
- Adaptive Reservoir Computing for Multi-Scenario Chaotic System Forecasting — Shadmehr Zaregarizi, Khashayar Yavari
- LSTM Variants for Chaotic Dynamical Systems: An Empirical Study on the Lorenz Attractor — Ruslan Gokhman
- A Metric-Adaptive Hybrid Pipeline for Chaotic Lorenz Dynamics in the Common Task Framework for Scientific Machine Learning — Dumitru Novic
🏢 Smart Buildings Challenge
Challenge Papers
- PIRS: Physics-Informed Reward Shaping for SAC-Based Building Energy Management — Shadmehr Zaregarizi, Khashayar Yavari
- Quantifying the Energy Floor: Direct Measurement and Replay Buffer Bias in SAC-Based HVAC Control on sbsim — Bo Li, Chen Zhang
- Tensorized Radiative Heat Transfer for a Scalable and Calibrated Building Energy Simulator — Sang woo Ham, Donghun Kim, Michael Rossetti, John Sipple
Speakers
Michael Rossetti
Google Smart Buildings Simulator: An Open Source Research Project
Vincent Taboga
Mila
AI for HVAC control at scale: learning controllers that adapt across buildings
Matt Levine
Basis / MIT
Dynestyx: Modernizing Bayesian tools for dynamical systems
Alexey Yermakov
University of Washington
The Common Task Framework for Scientific Machine Learning
Baosen Zhang
University of Washington
Dynamic Control and Integration of Data Centers into the Grid
Schedule
Call for Papers
Submissions are now closed.
Submission Requirements
- Maximum Length: 4 pages (excluding references). Camera-ready allows 5 pages.
- Appendices permitted but not required for reviewers to consider
- Use ACM conference template
- Double-blind review process
- Submit here: aideeds26.hotcrp.com
Subject Areas
Physics-Informed ML
- Integrating physical constraints into models
- Regularization via physics priors
- Symmetry-aware architectures
- Real-time control using physics-based ML
Hybrid Physics & Data-Driven Models
- Physical constraints with data-driven time-series
- Reduced-order modeling for complex flows
- Discovering governing equations
- Integrating data collection, control, modeling
Optimization in Dynamical Systems
- Nonconvex optimization with physical constraints
- Sensor optimization and uncertainty quantification
- Optimal sensor placement
- Control theory integration with ML
Reinforcement Learning for Control
- Scalable RL for multi-agent systems
- Robust RL under model mismatch
- Model-based and model-free hybrid control
- Physics-driven RL architectures
Uncertainty & Risk Quantification
- Uncertainty in sensor placement and control
- Robust control under uncertainty
- Ethical and interpretable models for safety
- Real-time control with embedded uncertainty
Ethics & Policies
All submissions must adhere to ACM e-Energy Policies and the ACM Code of Ethics. LLMs are permitted as tools, but authors are responsible for content. We do not accept LLMs as authors. Dual submissions are allowed but novel submissions receive higher priority.
Organizers
Judah Goldfeder
Columbia University
Main Contact jag2396@columbia.edu
PhD Candidate and research fellow at NSF AI Institute in Dynamic Systems
Philippe Wyder
Distyl AI
Applied AI Researcher
Na (Lina) Li
Harvard University
Winokur Family Professor of Electrical Engineering and Applied Mathematics
Bing Dong
Syracuse University
Research on smart buildings and human-building interactions
Alexey Yermakov
University of Washington
Researcher in scientific ML, physics-guided AI, and dynamical systems
Hod Lipson
Columbia University
Professor of Engineering and Data Science, Chair of Mechanical Engineering
J. Nathan Kutz
University of Washington
Director, NSF AI Institute in Dynamic Systems
Jiong Lin
Columbia University
Department of Mechanical Engineering
Acknowledgments
This workshop is supported by the NSF AI Institute in Dynamic Systems (Grant #2112085).
