Research theme
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 the research mapA 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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.