New algorithms reduce dynamic regret in non-stationary RL environments.
problem Optimizing policies in environments that change over time.
method POWER and POWER++ algorithms for policy optimization with dynamic regret analysis.
result POWER++ improves dynamic regret by actively adapting to non-stationarity.
Proposes a method to learn policies from offline data with reduced bias.
problem Learning policies from offline data with reduced bias and complexity constraints.
method Cross-fitted debiasing device for policy learning from offline data.
result Achieves N \sqrt N N regret for complex policy classes with a product-of-errors nuisance remainder. New algorithm reduces decision switching in dynamic environments.
problem Online learning with memory and non-stationary environments.
method Dynamic policy regret, novel ensemble approach, meta-base decomposition.
result Proves optimal dynamic policy regret for memory length, non-stationarity, and time horizon.
New model-free algorithm achieves similar LQR regret guarantees.
problem Model-free control of linear dynamical systems under quadratic costs.
method Online policy gradient scheme with policy space cost analysis.
result Achieves regret scaling with √T, matching model-based methods.
Study online control of unknown time-varying systems with negative and positive results.
problem Online control of time-varying systems with unknown dynamics.
method Algorithmic upper bounds and lower bounds for different policy classes.
result Sublinear adaptive regret bounds for Disturbance Response policies.
A method to minimize regret in multi-agent control systems with adversarial disturbances.
problem Optimal control of dynamical systems with adversarial disturbances and multiple agents.
method Reduction from online convex optimization to a distributed algorithm for multi-agent control.
result The resulting distributed algorithm has low regret relative to the optimal precomputed joint policy.
We study the dynamic assortment planning problem, where for each arriving customer, the seller offers an assortment of substitutable products and customer makes the purchase among offered products according to an uncapacitated multinomial logit (MNL) model. Since all the utility parameters of MNL are unknown, the selle…
New algorithm minimizes worst-case regret in uncertain, time-varying dynamics.
problem Model-based policy learning in uncertain, time-varying dynamics.
method Planning regret metric and iterative algorithm for minimizing it.
result Empirical evidence shows the proposed algorithm outperforms existing methods.
New algorithms reduce dynamic regret in online MDPs with changing losses.
problem Online MDPs with adversarial loss changes and known transitions.
method Dynamic regret measure, novel ensemble algorithms for three models.
result Provably optimal dynamic regret bounds for episodic SSP, improved bounds for predictable environments.
Study strategic dynamic pricing for buyers with unknown manipulation costs.
problem Strategic buyers manipulate their features to get lower prices, hindering profit maximization.
method Proposes a strategic dynamic pricing policy that incorporates strategic behavior and binary response data.
result Achieves sublinear regret bound of O ( T ) O(\sqrt{T}) O ( T ) compared to linear Ω ( T ) Ω(T) Ω ( T ) regret of non-strategic policies. Dynamic pricing policy converges to Nash equilibrium with low regret.
problem Sequential price competition among sellers over multiple periods.
method Semi-parametric least-squares estimation of s-concave demand functions.
result Prices converge to Nash equilibrium with rate O ( T − 1 / 7 ) O(T^{-1/7}) O ( T − 1/7 ) and sellers incur regret O ( T 5 / 7 ) O(T^{5/7}) O ( T 5/7 ) . Study agnostic feature-based dynamic pricing models with linear policies and noisy valuations.
problem Tackles dynamic pricing with unknown noise and no assumptions on data.
method Studies two agnostic models: linear policy and linear noisy valuation, presenting algorithms and regret bounds.
result Demonstrates no-regret learning is possible under weak assumptions, but noisy feedback is not significantly more useful than bandit feedback.
An important problem in sequential decision-making under uncertainty is to use limited data to compute a safe policy, i.e., a policy that is guaranteed to perform at least as well as a given baseline strategy. In this paper, we develop and analyze a new model-based approach to compute a safe policy when we have access …
New algorithm reduces dynamic regret for MDPs with unknown transition and adversarial rewards.
problem Episodic linear mixture MDPs with unknown transition and adversarial rewards.
method Combines occupancy-measure-based global optimization and policy-based variance-aware value-targeted regression.
result Achieves near-optimal dynamic regret of O ~ ( d H 3 K + H K ( H + P ˉ K ) ) \widetilde{\mathcal{O}}(d \sqrt{H^3 K} + \sqrt{HK(H + \bar{P}_K)}) O ( d H 3 K + H K ( H + P ˉ K ) ) . The paper addresses fairness in dynamic pricing for strategic buyers.
problem Price disparities among specific groups can lead to unfair perceptions and legal violations.
method Proposes a dynamic pricing policy that achieves fairness and discourages strategic behavior.
result Achieves an upper bound of O ( T + H ( T ) ) O(\sqrt{T}+H(T)) O ( T + H ( T )) regret over T T T time horizons, reducing regret by 35.06% compared to a benchmark policy. Study minimax rates for online learning with time-varying dynamics.
problem Online learning with time-varying state and cost dynamics.
method Non-constructive upper and lower bounds, complexity and stability terms.
result Characterization of minimax rates and necessary conditions for learnability.
Paper develops a fair pricing algorithm for dynamic settings with uncertain demand.
problem Fair pricing in dynamic, uncertain demand scenarios.
method Contextual bandit algorithm with dynamic pricing and demand learning.
result Achieves optimal regret bound with fairness constraints.
New algorithm tackles dynamic query routing to multiple embedding models.
problem Dynamic query routing to multiple embedding models under adversarial conditions.
method Formalized as adversarial contextual linear bandit with low-rank experts, proposed HPG algorithm.
result HPG algorithm achieves linearized policy regret of i l d e O ( s M T ) ilde{\mathcal O}(s\sqrt{M T}) i l d e O ( s M T ) . Develops methods for dynamic pricing in incomplete data settings.
problem Incomplete historical data makes optimal pricing difficult.
method Nonparametric partial identification framework for offline dynamic pricing.
result Pessimistic and opportunistic policies with regret bounds.
Study on learning strategies in adaptive Markov games with policy regret as metric.
problem Learning in dynamic Markov games with adaptive opponents is challenging.
method Introduced policy regret as a new learning metric and developed algorithms for consistent adaptive adversaries.
result Achieved T \sqrt{T} T policy regret against certain adaptive adversaries. Optimal algorithm for LQR control with improved regret bound.
problem Nonstochastic control with quadratic losses (LQR control).
method Online algorithm with optimal dynamic regret of i l d e O ( e x t m a x { n 1 / 3 T V ( M 1 : n ) 2 / 3 , 1 } ) ilde{O}( ext{max}\{n^{1/3} \mathcal{TV}(M_{1:n})^{2/3}, 1\}) i l d e O ( e x t ma x { n 1/3 TV ( M 1 : n ) 2/3 , 1 }) . result Improves the best known rate of i l d e O ( n ( T V ( M 1 : n ) + 1 ) ) ilde{O}(\sqrt{n (\mathcal{TV}(M_{1:n})+1)} ) i l d e O ( n ( TV ( M 1 : n ) + 1 ) ) for general convex losses. Study optimal policy regret in partially observable Markov games with adaptive opponents.
problem Optimal sequential decision-making in partially observable environments against strategic, adaptive opponents.
method An epoch-based optimistic maximum-likelihood algorithm that selects one policy per epoch using confidence sets built cumulatively from past data.
result Achieves i l d e O ( T ) ilde{O}(\sqrt{T}) i l d e O ( T ) policy regret for fixed problem parameters, with explicit dependence on horizon, adversary memory, confidence radius, and aggregate Eluder dimension. Doubly fair dynamic pricing ensures equal prices for different groups over time.
problem Achieving equal prices for different groups in online dynamic pricing.
method Online learning algorithm that balances procedural and substantive fairness.
result Achieves i l d e O ( T ) ilde{O}(\sqrt{T}) i l d e O ( T ) regret, zero procedural unfairness, and i l d e O ( T ) ilde{O}(\sqrt{T}) i l d e O ( T ) substantive unfairness. New RL algorithm tackles non-stationary environments with flexible policy updates.
problem Non-stationary reinforcement learning with time-varying rewards and transition probabilities.
method Model-free policy-based algorithm NS-NAC with restart-based exploration and dynamic learning rates.
result Dynamic regret of i l d e O ( ∣ S ∣ 1 / 2 ∣ A ∣ 1 / 2 Δ T 1 / 6 T 5 / 6 ) ilde{\mathscr O}(|S|^{1/2}|A|^{1/2}Δ_T^{1/6}T^{5/6}) i l d e O ( ∣ S ∣ 1/2 ∣ A ∣ 1/2 Δ T 1/6 T 5/6 ) for both algorithms. New algorithm reduces control error in systems with changing dynamics.
problem Online control of systems with time-varying linear dynamics.
method Introduces adaptive regret metric and a novel meta-algorithm.
result First adaptive regret bound for online convex optimization with memory.
Study non-rectangular robust MDPs for average-reward, finding optimal policies and transient values.
problem Non-rectangular robust Markov decision processes under average-reward criterion.
method Proves history-dependent policies are robust-optimal, introduces transient-value framework, constructs epoch-based policy.
result Existence and properties of robust optimal policies, transient value bounds.
Supplier learns to price contracts against a learning retailer.
problem Designing data-driven pricing policies for a supplier facing a learning retailer.
method Connecting to non-stationary online learning, proposing dynamic pricing policies for discrete and continuous demand.
result Supplier's pricing policies lead to sublinear regret bounds under various retailer learning policies.
PCGS-TF uses a Transformer to adaptively control expert switching in non-stationary environments.
problem Static regret is insufficient for strictly online prediction in non-stationary settings.
method Policy-Controlled Generalized Share (PCGS) with a Transformer as an update controller.
result PCGS-TF achieves the lowest dynamic regret in non-stationary families and expert pools.
New algorithm reduces dynamic regret for noisy gradient feedback with piecewise polynomial comparators.
problem Online estimation of piecewise polynomial trends with noisy feedback.
method Introduces variational constraint for piecewise polynomial comparators, designs adaptive algorithm.
result Achieves nearly optimal dynamic regret of $ ilde{O}(n^{ rac{1}{2k+3}}C_n^{ rac{2}{2k+3}})$ .
Study dynamic pricing with semi-parametric models to minimize regret.
problem Optimizing dynamic pricing in a noisy market with binary sales outcomes.
method Proposes a semi-parametric statistical learning policy combining GLM and online decision-making.
result Achieves a regret upper bound of $ ilde{O}_{d}(T^{ rac{2m+1}{4m-1}})$ under mild conditions.
We consider a multi-armed bandit problem in a setting where each arm produces a noisy reward realization which depends on an observable random covariate. As opposed to the traditional static multi-armed bandit problem, this setting allows for dynamically changing rewards that better describe applications where side inf…
Study nonparametric contextual bandits with batched updates, achieving optimal regret.
problem Optimal regret in nonparametric contextual bandits with batch constraints.
method Dynamic binning of covariate space, optimal regret achieved.
result Achieves optimal regret (up to logarithmic factors) for nonparametric contextual bandits.
Most contextual bandit algorithms minimize regret against the best fixed policy, a questionable benchmark for non-stationary environments that are ubiquitous in applications. In this work, we develop several efficient contextual bandit algorithms for non-stationary environments by equipping existing methods for i.i.d. …
Paper presents a privacy-preserving method for dynamic assortment selection.
problem Personalized assortment recommendations with data privacy concerns.
method Perturbed upper confidence bound method integrating calibrated noise.
result Policy satisfies Joint Differential Privacy (JDP) with near-optimal regret bound.
Evaluating AI investment strategies
problem Auditing a black-box algorithmic decision-maker
method Exact decomposition of cumulative regret
result Cumulative regret equals sum of per-period covariances
Study learns optimal bidding strategy in auctions with dynamic values and aggregated feedback.
problem Optimizing bidding in auctions with time-dependent values and limited feedback.
method Combines plug-in estimators with differential-equation characterization of optimal policy.
result Achieves near optimal regret bounds for learning optimal policy.
Paper shows re-solving heuristics have constant regret for price-based revenue management.
problem Optimal pricing policies for revenue management with time constraints.
method Proves re-solving heuristics have O ( 1 ) O(1) O ( 1 ) regret compared to optimal policies. result Improved regret bound to O ( 1 ) O(1) O ( 1 ) from O ( ln T ) O(\ln T) O ( ln T ) , complemented by Ω ( ln T ) Ω(\ln T) Ω ( ln T ) gap with fluid model. Paper combines RL with policy regularization for inventory policies.
problem Optimizing inventory policies using RL and dynamic programming.
method Hybrid approach combining RL with policy regularization.
result Generalization guarantees for inventory policies using VC theory.
New control methods for systems with adversarial perturbations.
problem Control systems with adversarial noise.
method Online convex optimization and convex relaxations.
result Low regret policies against adversarial perturbations.
New algorithm adapts to unknown demand smoothness for dynamic pricing.
problem Dynamic pricing with unknown Hölder smoothness of demand function.
method Self-similarity condition and adaptive algorithm.
result Adaptive algorithm achieves minimax optimal regret without prior knowledge of smoothness.
Study shows how to learn optimal policies quickly in stochastic control problems.
problem Learning optimal policies in large, continuous state and action spaces with limited data.
method Analyzes three geometric exponents to quantify fast policy regret convergence.
result Shows that fast policy regret convergence is induced by specific geometric structures.
Paper develops privacy-preserving dynamic pricing policy for e-commerce.
problem Protecting customer privacy in dynamic pricing with personalized information.
method Uses differential privacy framework to develop a privacy-preserving policy.
result Achieves both privacy and performance guarantees in dynamic pricing.
We consider the problem of multi-product dynamic pricing, in a contextual setting, for a seller of differentiated products. In this environment, the customers arrive over time and products are described by high-dimensional feature vectors. Each customer chooses a product according to the widely used Multinomial Logit (…
Improved RL algorithm stabilizes unknown linear systems with polynomial regret.
problem Learning and stabilizing unknown linear dynamical systems.
method Proposes an algorithm with an improved exploration strategy for fast stabilization.
result Achieves i l d e O ( T ) ilde{\mathcal{O}}(\sqrt{T}) i l d e O ( T ) regret after T T T time steps. Improved reinforcement learning algorithm with linear approximation for unknown dynamics.
problem Reinforcement learning with adversarial changing cost functions and bandit feedback.
method Combines mirror-descent and least squares policy evaluation in an auxiliary MDP.
result Obtains an O ~ ( K 6 / 7 ) \widetilde O(K^{6/7}) O ( K 6/7 ) regret bound, significantly improving over previous methods. The notion of \emph{policy regret} in online learning is a well defined? performance measure for the common scenario of adaptive adversaries, which more traditional quantities such as external regret do not take into account. We revisit the notion of policy regret and first show that there are online learning settings …
EPIC quantifies reward differences without policy optimization.
problem Distinguishing reward function quality from policy optimization issues.
method EPIC distance to compare reward functions directly.
result EPIC bounds policy training success and regret.
Motivated by pricing in ad exchange markets, we consider the problem of robust learning of reserve prices against strategic buyers in repeated contextual second-price auctions. Buyers' valuations for an item depend on the context that describes the item. However, the seller is not aware of the relationship between the …