Turning context into control
- Instructions
- Predictions
- Control
Human instructions become disturbance predictions, connecting language models to feedback control.
Research
Reliable decisions in an uncertain world.
I study how learning improves decisions under imperfect predictions and models, connecting mathematical guarantees, AI agents, and energy infrastructure.
Research landscape
Learning with predictionsWhen should a decision maker trust a prediction?
Safe and distributed controlHow can a controller learn while preserving reliable operation?
Language and decision agentsHow can language and world models support dependable decisions?
Energy and infrastructureHow can intelligent infrastructure use energy more effectively?
Strategic and equitable learningWhat changes when decisions affect other decision makers?
Information and network learningWhat information and structure are needed to learn reliably?
Good predictions create opportunities. Reliable algorithms make them useful.
Information, trust, and physical constraints connect these research themes.Learning with predictions
Algorithms that benefit from accurate advice while limiting the cost of errors. This thread studies consistency, robustness, competitive analysis, and the value of different forms of predictive information.
Explore this themeSafe and distributed control
Predictive control with performance and feasibility guarantees. The work spans perturbation analysis, nonlinear policy certification, aggregate flexibility, and distributed system level synthesis.
Explore this themeLanguage and decision agents
Language models provide context, preferences, and predictions for sequential decisions. Algorithmic guidance and structural certification connect these capabilities to explicit performance guarantees.
Explore this themeEnergy and infrastructure
Control and learning for electric vehicle charging, renewable integration, demand forecasting, water systems, and grid coordination. Physical constraints and changing operating conditions shape the algorithms.
Explore this themeStrategic and equitable learning
Learning and optimization with interacting agents. This work examines uncertain beliefs, strategic advice, equilibrium, and the distribution of benefits from shared predictive models.
Explore this themeInformation and network learning
Learning network structure from limited measurements and adapting graph models to heterogeneous local geometry. These results connect measurement complexity, physical networks, and useful representations.
Explore this themeSelected projects
Browse papersLearning what to trust
- Mixed signals
- Separation
- Trust adaptation
Separate mixed disturbances and adapt confidence in each prediction source as conditions change.
A geometry for every node
- Network
- Local geometry
- Graph learning
Graph learning that adapts to local structure instead of imposing one geometry on an entire network.
Energy systems in context
- Microgrid
- Context and data
- Simulation
An open platform linking physical energy measurements, human context, and simulation for energy management.
Research papers
2026AAAI
Adaptive Riemannian Graph Neural Networks
ARGNN learns a node specific Riemannian metric instead of imposing one fixed curvature on an entire graph. A diagonal metric parameterization and geometric regularization support efficient training, with convergence analysis and experiments on homophilic and heterophilic graphs.
2026Applied Energy
Counterfactual load forecasting with LLM-structured events and representation learning
NACF turns news into structured treatments and estimates load trajectories under alternative event conditions. Reweighting and representation balancing address observed confounding. Experiments examine factual accuracy and interpretable demand perturbations without claiming that unobserved counterfactual outcomes can be directly validated.
2026ACM e-Energy
Energy Injection Identification enabled Disaggregation with Deep Multi-Task Learning
DualNILM jointly recognizes appliance states and identifies energy injected behind the meter. Its transformer architecture combines temporal learning tasks to separate consumption from injections, with evaluation on measured and synthesized datasets.
2026AISTATS
Leveraging Machine-Learned Advice in Strategic Interactions with No-Regret Learners
A measure of advice quality connects simulators and payoff predictions to strategic performance. The paper establishes benefits of reliable advice for approximate Stackelberg play and limitations on simultaneously exploiting accurate advice and protecting against inaccurate advice.
2026IEEE PES International Meeting
PEARL: A Physics-Enhanced Adaptive Residual Learning Framework for PV Modeling
PEARL separates a physical photovoltaic simulator from a learned residual correction. A controlled benchmark compares several machine learning architectures and a coupled modeling baseline, examining accuracy while retaining the physical model as an interpretable reference.
2026POMACS
Prediction-Specific Design of Learning-Augmented Algorithms
Strong optimality refines the usual consistency and robustness tradeoff to account for each prediction. A bilevel optimization framework constructs algorithms for ski rental and one max search, with case studies in power management and trading.
2026CSEE JPES
Quantum Learning and Estimation for Coordinated Operation between Distribution Networks and Energy Communities
A hybrid quantum learning model estimates energy community responses to prices, while quantum amplitude estimation supports uncertainty calculations. Numerical studies assess accuracy and resource requirements. Reported timing advantages assume ideal quantum devices rather than deployed hardware.
2026IEEE Control Systems Letters
VigilMPC: Certified Online Tube Model Predictive Control for Safe Real-Time Neuromodulation
VigilMPC accepts a predictor update only when a backup plan remains feasible under tightened constraints. Recursive feasibility and constraint satisfaction follow under the tube MPC assumptions. Evaluation uses simulated neural dynamics and a switching benchmark, rather than clinical deployment.
2026ICML
World Models in Pieces: Structural Certification for General Agents
Structural certification links performance on compositional goals to local guarantees on world model transitions. The results identify reliable pieces of a model without requiring universal competence, and establish limits on the precision of such guarantees.
2025ACL Findings
Beyond Numeric Rewards: In-Context Dueling Bandits with LLM Agents
The study identifies a gap between quick preference discovery and sustained exploitation by language agents. LEAD combines LLM reasoning with dueling bandit algorithms to obtain weak and strong regret guarantees, with evaluations under noisy and adversarial prompts.
2025IEEE Transactions on Transportation Electrification
Hierarchical Game for Coupled Power System With Energy Sharing and Transportation System
A hierarchical game couples an energy sharing mechanism with transportation decisions and power network constraints. The paper establishes properties of the energy sharing equilibrium and reformulates the coupled problem for numerical solution.
2025CDC
INSTRUCT MPC: A Human-LLM-in-the-Loop Framework for Context-Aware Control
A language to distribution module converts contextual instructions into disturbance predictions for MPC. The framework closes the loop between instructions, predictions, and control, with a regret analysis for linear dynamics under the stated training assumptions.
2025ACM e-Energy
Learning-Augmented Online Control for Decarbonizing Water Infrastructures
LAOC restricts online decisions to safe action sets while exploiting learned predictions for pump scheduling. Analysis characterizes the relationship between safety requirements and operating costs, with a building water supply case study.
2025ACM e-Energy demo
Open In-Context Energy Management Platform
OpenCEM presents a platform design connecting energy time series, events, human context, and simulation. An on campus solar and battery installation motivates evaluation of context sensitive control. The paper describes planned data and API capabilities.
2025IEEE Transactions on Transportation Electrification
Out-of-Distribution-Aware Electric Vehicle Charging
OOD Charging adapts an awareness radius using temporal difference errors to balance learned scheduling with a robust baseline. Evaluation on ACN data examines behavioral changes during COVID 19 and the resulting distribution shift.
2025CDC
PredSLS: A System-Level Framework for Distributed Predictive Control
PredSLS integrates forecasts directly into system level synthesis with locality constraints. Temporal decay of prediction effects supports finite horizon approximations, while spatial and temporal decomposition enable scalable computation across a network.
2025NeurIPS
Reinforcement Learning with Imperfect Transition Predictions: A Bellman-Jensen Approach
A Bayesian value function and Bellman Jensen gap quantify the value of imperfect transition forecasts. BOLA separates offline value learning from online adaptation, with sample efficiency analysis and experiments in synthetic environments and wind energy storage control.
2025Environmental Data Science
Robust machine-learned algorithms for efficient grid operation
RobustML combines a learned dispatch algorithm with a bound on worst case cost. A combined cycle cogeneration case study examines dispatch and ramping costs as renewable penetration increases, including performance under distribution shift.
2025IEEE Energy Sustainability Magazine
Robust Solar Hosting Capacity for Pasadena's Decarbonization: A Mathematical Program for Robustness Against Uncertainty
The Pasadena case study formulates solar hosting capacity planning as a robust optimization problem. It connects uncertainty in renewable generation and demand to distribution network constraints and practical choices along a municipal decarbonization pathway.
2024IEEE Transactions on Industrial Informatics
A Data-Driven Approach for Optimizing Early-Stage Electric Vehicle Charging Station Placement
Two proximity models represent user satisfaction with station locations. Population, travel, road network, and point of interest data inform a submodular optimization problem, solved approximately with a greedy method and evaluated in Haikou.
2024ICML
Building Socially-Equitable Public Models
The Equitable Objective trains a public predictive model around the outcomes of heterogeneous downstream agents. A policy gradient method and theoretical analysis study a more equitable distribution of performance, beyond optimizing prediction accuracy alone.
2024NeurIPS
Disentangling Linear Quadratic Control with Untrusted ML Predictions
DISC learns confidence in predictions of latent disturbance components. Competitive analysis covers linear and more general mixing functions, showing how online confidence adaptation can exploit accurate forecasts while maintaining protection against large errors.
2024NeurIPS
Safe Exploitative Play with Untrusted Type Beliefs
The paper formalizes a risk and opportunity tradeoff for agents with uncertain beliefs about other players. Upper and lower bounds characterize the Pareto frontier in normal form and stochastic Bayesian games, with numerical illustrations of the tradeoff.
2023NeurIPS
Anytime-Competitive Reinforcement Learning with Policy Prior
ACRL constrains cumulative costs relative to a policy prior at each round, rather than only in expectation over episodes. The analysis establishes cost guarantees and regret relative to the constrained optimum, with experiments in carbon aware computing.
2023NeurIPS
Beyond Black-Box Advice: Learning-Augmented Algorithms for MDPs with Q-Value Predictions
Q value predictions expose more information than an opaque policy recommendation. For a single trajectory MDP, the analysis characterizes consistency and robustness tradeoffs and shows how structured advice improves the guarantees available from a robust baseline.
2023IEEE Open Journal of Control Systems
Certifying Black-Box Policies With Stability for Nonlinear Control
Simply blending two stabilizing policies can cause instability. An adaptive confidence policy instead uses approximate model information to certify stability, with competitive guarantees under bounded nonlinearity and evaluations on Cart Pole and charging under distribution shift.
2022NeurIPS
Bounded-Regret MPC via Perturbation Analysis: Prediction Error, Constraints, and Nonlinearity
A general analysis pipeline converts perturbation bounds for finite horizon control into dynamic regret bounds for MPC. It handles prediction errors in costs, dynamics, and disturbances, and extends the analysis to constrained and nonlinear settings under suitable regularity conditions.
2022POMACS
Robustness and Consistency in Linear Quadratic Control with Untrusted Predictions
A confidence parameter governs how much a linear quadratic controller trusts disturbance predictions. Competitive bounds describe the tradeoff between accurate and inaccurate advice, and a self tuning policy adapts confidence online using observed prediction quality.
2022Electric Power Systems Research
Towards balanced three-phase charging: Phase optimization in adaptive charging networks
PXA solves a relaxation of the phase optimization problem, with convergence guarantees under specified conditions. Its MPC implementation updates decisions online. Experiments use adaptive charging network data to examine energy delivery, charging costs, and phase balance.
2021SmartGridComm
Data-driven Electric Vehicle Charging Station Placement for Incentivizing Potential Demand
A proximity based satisfaction model turns station placement into monotone submodular maximization. A greedy algorithm has a provable approximation ratio, with a Haikou case study using population, travel, and point of interest data.
2021IEEE Transactions on Power Systems
Decentralized Provision of Renewable Predictions Within a Virtual Power Plant
A decentralized prediction provision scheme lets consumers obtain local forecasts and exchange summaries with a virtual power plant. The analysis studies social surplus, convergence, and the demand gap relative to centralized information sharing.
2021POMACS
Information Aggregation for Constrained Online Control
Entropic aggregation summarizes a local controller's feasible actions for a remote decision maker. Penalized Predictive Control can approach the regret of full information control under causal invariance and a sufficiently long prediction window.
2021IEEE Transactions on Smart Grid
Learning-Based Predictive Control via Real-Time Aggregate Flexibility
Maximum entropy feedback summarizes the feasible actions of controllable loads. Reinforcement learning approximates this feedback for Penalized Predictive Control, reducing information and computation requirements. Charging data demonstrates the coordination approach.
2020Electric Power Systems Research
Classification of electric vehicle charging time series with selective clustering
An iterative procedure extracts and clusters the tail portions of charging curves despite missing observations, variable lengths, scheduling effects, and measurement noise. Experiments on ACN data illustrate how these patterns can support useful charging models.
2020POMACS
Competitive Algorithms for the Online Multiple Knapsack Problem with Application to Electric Vehicle Charging
A general fractional multiple knapsack model captures assignment and rate constraints. An online primal dual algorithm obtains near optimal competitive guarantees and is evaluated using charging traces. This work provides an online allocation foundation without requiring learned predictions.
2020IEEE TSIPN
Learning Graphs From Linear Measurements: Fundamental Trade-Offs and Applications
Bounds for noisy and noiseless graph recovery connect sparsity, graph distributions, and linear measurements. A three stage recovery scheme and practical algorithm are studied on canonical graphs and electrical grid admittance matrices.
2020ACM e-Energy
ORC: An Online Competitive Algorithm for Recommendation and Charging Schedule in Electric Vehicle Charging Network
ORC jointly recommends stations and prices while scheduling charging for sequential arrivals. A customizable value function supports competitive guarantees despite uncertainty in arrivals and user choices. Experiments compare operator revenue with benchmark scheduling methods.
2019CDC
Learning Graph Parameters from Linear Measurements: Fundamental Trade-offs and Application to Electric Grids
Information theoretic bounds relate measurement requirements to graph complexity and recovery error. A recovery algorithm estimates topology and parameters from linear measurements, with demonstrations on graph families and electrical grid test cases.
No matching papers
Try a different phrase, theme, or year.
Connections across the work
Conceptual links between research themes, grounded in the papers below.
From prediction error to control performance
Trust adaptation in linear quadratic control and perturbation bounds for MPC connect the quality of a forecast to the cost of acting on it. Nonlinear certification adds stability to this picture.
- Robustness and Consistency in Linear Quadratic Control with Untrusted PredictionsPOMACS 2022
- Bounded-Regret MPC via Perturbation Analysis: Prediction Error, Constraints, and NonlinearityNeurIPS 2022
- Certifying Black-Box Policies With Stability for Nonlinear ControlIEEE Open Journal of Control Systems 2023
- Disentangling Linear Quadratic Control with Untrusted ML PredictionsNeurIPS 2024
From language context to constrained actions
INSTRUCT MPC translates human instructions into disturbance predictions. World model certification and VigilMPC study complementary ways of deciding which learned models can support reliable planning and safe updates.
Algorithms guide learned decision makers
Q value advice and imperfect transition forecasts provide structured information for reinforcement learning. LEAD applies a related principle by supporting language agents with dueling bandit algorithms.
Guarantees meet physical infrastructure
Aggregate flexibility communicates feasible actions for coordinated charging. Robust dispatch and safe water control connect performance guarantees to operational costs and physical constraints.
- Learning-Based Predictive Control via Real-Time Aggregate FlexibilityIEEE Transactions on Smart Grid 2021
- Information Aggregation for Constrained Online ControlPOMACS 2021
- Robust machine-learned algorithms for efficient grid operationEnvironmental Data Science 2025
- Learning-Augmented Online Control for Decarbonizing Water InfrastructuresACM e-Energy 2025
Context becomes an energy system input
OpenCEM motivates shared contextual energy benchmarks. INSTRUCT MPC uses instructions for control, while counterfactual load forecasting represents news events as interventions on electricity demand.
- Open In-Context Energy Management PlatformACM e-Energy demo 2025
- INSTRUCT MPC: A Human-LLM-in-the-Loop Framework for Context-Aware ControlCDC 2025
- Counterfactual load forecasting with LLM-structured events and representation learningApplied Energy 2026
Prediction errors become strategic risks
In strategic environments, advice quality affects both opportunity and risk. The Bayesian game and repeated game results expose limits on combining exploitation with protection against inaccurate beliefs.
Coordination through information and incentives
Virtual power plants use renewable predictions to coordinate consumers. Energy sharing and transportation games model how charging demand, prices, and network constraints interact.
- Decentralized Provision of Renewable Predictions Within a Virtual Power PlantIEEE Transactions on Power Systems 2021
- Hierarchical Game for Coupled Power System With Energy Sharing and Transportation SystemIEEE Transactions on Transportation Electrification 2025
- Building Socially-Equitable Public ModelsICML 2024
Recovering structure from limited measurements
Graph recovery estimates electrical network parameters. Charging curve clustering and energy disaggregation recover useful structure from observed behavior and aggregate meter readings.
- Learning Graphs From Linear Measurements: Fundamental Trade-Offs and ApplicationsIEEE TSIPN 2020
- Classification of electric vehicle charging time series with selective clusteringElectric Power Systems Research 2020
- Energy Injection Identification enabled Disaggregation with Deep Multi-Task LearningACM e-Energy 2026
Open research directions
Questions motivated by this body of work. These are directions to explore, rather than established results.
Open question
Certified contextual control
Can contextual and world model updates be certified before they affect a physical system?
Connect language conditioned predictions with local world model guarantees and feasibility checks. An open challenge is preserving safety when both the context and the learned dynamics change during operation.
Starting points
Open question
Prediction quality as a decision resource
Which prediction errors actually matter for the decision being made?
Move from a single forecast accuracy score to guarantees that reflect the prediction, the task, and the affected transitions. This suggests adaptive confidence allocation across sources, horizons, and latent disturbances.
Starting points
- Prediction-Specific Design of Learning-Augmented AlgorithmsPOMACS 2026
- Disentangling Linear Quadratic Control with Untrusted ML PredictionsNeurIPS 2024
- Beyond Black-Box Advice: Learning-Augmented Algorithms for MDPs with Q-Value PredictionsNeurIPS 2023
- Reinforcement Learning with Imperfect Transition Predictions: A Bellman-Jensen ApproachNeurIPS 2025
Open question
Coordinated low carbon infrastructure
Can distributed energy and water resources adapt together under changing conditions?
Combine local feasibility summaries with distributed predictive control and contextual energy models. Open questions include communication limits, distribution shift, and guarantees across coupled physical networks.
Starting points
- PredSLS: A System-Level Framework for Distributed Predictive ControlCDC 2025
- Learning-Based Predictive Control via Real-Time Aggregate FlexibilityIEEE Transactions on Smart Grid 2021
- Out-of-Distribution-Aware Electric Vehicle ChargingIEEE Transactions on Transportation Electrification 2025
- Learning-Augmented Online Control for Decarbonizing Water InfrastructuresACM e-Energy 2025
- Open In-Context Energy Management PlatformACM e-Energy demo 2025
Open question
Accountable learning across agents
How should shared models balance efficiency, strategic risk, and equity?
Connect the downstream objectives of public models to robust strategic decision making. A key challenge is evaluating advice when its benefits and failures are distributed unevenly across participating agents.