Value of information
Fundamental limits for exact preview, delayed preview, finite-horizon forecasts, and contextual side information under H2, H-infinity, regret, competitive ratio, robustness, and sample-complexity metrics.
IEEE Conference on Decision and Control 2026
Control under Unreliable Predictions, Previews, and Contextual Information
A half-day workshop on how feedback controllers should exploit forecasts, previews, learned context, and side information when the information itself can be delayed, biased, intermittent, shifted, or adversarially wrong.
Monday, December 14, 2026 / Half-day / Hilton Hawaiian Village, Honolulu
Workshop overview
When forecasts become part of the control loop
Preview and lookahead information have long been central to control, especially through model predictive control and its robust and stochastic variants. What is changing is the nature of the predictive information available to modern control systems.
Forecasts from weather models, traffic predictors, learned dynamics models, foundation models, grid-load predictors, and sensor-rich robotic systems are increasingly high-dimensional, contextual, and data-driven. They can be highly informative, but also delayed, biased, mis-calibrated, distribution-shifted, intermittent, or adversarially wrong.
The program is organized around synthesis and analysis questions that sit between classical preview control, online optimization, learning-augmented algorithms, and safety-critical cyber-physical systems.
Central question
How should controllers decide when to trust prediction?
The goal is not simply to optimize over a forecast horizon, but to develop synthesis and analysis tools that quantify the value of predictions, decide when to trust or ignore them, and guarantee control performance based on the properties of the available information.
Core technical topics
A control-theoretic agenda for unreliable information
Learning-augmented algorithms
Robustness and consistency tradeoffs with untrusted predictions, including competitive ratio and dynamic regret for algorithms with predictions on costs, disturbances, and context.
Networked prediction-enabled control
Local forecasts, neighbor information, and communication-limited previews in large-scale multi-agent systems such as power distribution networks.
Safety and verification
Predictions and high-level context with invariance, reachability, specification satisfaction, and certifiable fallback behavior.
Delay-aware and connectivity-enabled control
Designs that use shared information while accounting for communication delay, intermittency, packet drops, and human or physical response delays.
Computation and reliability
Confidence adaptation, forecast calibration, fallback policies, and benchmarks that distinguish useful prediction from fragile overfitting to a forecast model.
Prediction-rich applications
Data centers, water infrastructure, demand response, connected vehicles, renewable energy systems, adaptive buildings, and other domains where forecasts are useful but structurally imperfect.
Program
Half-day morning schedule
Each talk slot is structured as 28 minutes of presentation plus 7 minutes of Q&A. The closing panel will focus on open problems and cross-cutting themes identified during the talks.
Session I
Adam Wierman
Leveraging Untrusted Predictions in Multi-Agent Strategic Settings
Steven H. Low
Efficient Computation with Many Random Scenarios
Tongxin Li
Predictive Networked Control: From Offline Parametrization to Online Adaptation
Coffee break
Informal discussion and speaker interaction
Session II
Necmiye Ozay
Value of Predictions for Safety Control
Gabor Orosz
Predictions and Delays in Multiscale Control Systems
Panel discussion
Open discussion and audience Q&A with all speakers, chaired by the workshop organizers.
Invited speakers
Perspectives from safety, online control, energy, and mobility
Steven H. Low
California Institute of Technology
Efficient Computation with Many Random Scenarios
Talk abstract
Abstract will be added when finalized.
Bio
Frank J. Gilloon Professor of Computing and Mathematical Sciences and Electrical Engineering. His research addresses real-time optimization and control of large-scale networked systems, especially power grids and EV charging networks.
Necmiye Ozay
University of Michigan
Value of Predictions for Safety Control
Talk abstract
Abstract will be added when finalized.
Bio
Chen-Luan Faculty Development Professor in Electrical Engineering and Computer Science. Her work develops formal and data-driven methods for control synthesis and verification of hybrid and cyber-physical systems, including frameworks for quantifying the value of preview information in safety control.
Adam Wierman
California Institute of Technology
Leveraging Untrusted Predictions in Multi-Agent Strategic Settings
Talk abstract
Abstract will be added when finalized.
Bio
Carl F. Braun Professor of Computing and Mathematical Sciences. His research connects online optimization and adversarial control of linear dynamical systems with provable competitive ratio and regret guarantees.
Gabor Orosz
University of Michigan
Predictions and Delays in Multiscale Control Systems
Talk abstract
Abstract will be added when finalized.
Bio
Professor of Mechanical Engineering and Civil and Environmental Engineering. His work focuses on nonlinear dynamics and control of connected automated vehicles, including traffic prediction under communication delays and intermittent packet loss.
Tongxin Li
The Chinese University of Hong Kong, Shenzhen
Predictive Networked Control: From Offline Parametrization to Online Adaptation
Talk abstract
Abstract will be added when finalized.
Bio
Assistant Professor and Presidential Young Fellow in the School of Data Science. His research focuses on control, online algorithms, and AI in power systems.
Organizers
Workshop committee
Jing Yu
University of Washington
Bio
Assistant Professor in Electrical and Computer Engineering and affiliate faculty with the Clean Energy Institute. Her research studies control theory, machine learning, online decision making, and distributed algorithms for sustainable energy systems.
Tongxin Li
CUHK-Shenzhen
Bio
Assistant Professor and Presidential Young Fellow in the School of Data Science. He works on control, online algorithms, and AI in power systems.
Taylan Kargin
Massachusetts Institute of Technology
Bio
Postdoctoral Associate in the Laboratory for Information and Decision Systems. His research interests include control theory, optimization, and machine learning.
Expected outcomes
From rigorous limits to deployable systems
The workshop aims to strengthen dialogue between control theory, online optimization, networked systems, and application communities that often study related problems through different technical languages.
Synthesize what is currently understood about the value and limits of predictive information in control.
Identify open problems on when predictions help, when they hurt, and how controllers should adapt trust online while maintaining stability, safety, and robustness.
Build a stronger bridge between theory and applications in energy, transportation, robotics, and large-scale networked infrastructure.
Give students and early-career researchers a visible entry point into technically rich, application-motivated problems.