New algorithm achieves optimal regret in average reward MDPs without prior bias information.
problem Achieving optimal regret in average reward MDPs with computational efficiency and without prior bias information.
method Projective Mitigated Extended Value Iteration (PMEVI) to compute bias-constrained optimal policies efficiently.
result First tractable algorithm with minimax optimal regret of O ~ ( s p ( h ∗ ) S A T ) \widetilde{\mathrm{O}}(\sqrt{\mathrm{sp}(h^*) S A T}) O ( sp ( h ∗ ) S A T ) . Efficient algorithm for global optimization of multivariate Lipschitz functions.
problem Global optimization of multivariate Lipschitz continuous functions.
method Proposes an efficient minimax optimal algorithm using a predetermined query creation rule.
result Achieves an average regret bound of O ( L n T − 1 n ) O(L\sqrt{n}T^{-\frac{1}{n}}) O ( L n T − n 1 ) , minimax optimal. Improved exploration in factored average-reward MDPs reduces regret.
problem Minimizing regret in unknown Factored Markov Decision Processes (FMDPs).
method DBN-UCRL strategy, inspired by UCRL2, uses Bernstein-type confidence sets for individual elements of the transition function.
result Achieves a regret bound with a leading term strictly improving over existing bounds.
A new model-free algorithm achieves near-optimal regret for infinite-horizon MDPs.
problem Model-free reinforcement learning for infinite-horizon average-reward MDPs.
method Exploration Enhanced Q-learning (EE-QL) for weakly communicating MDPs.
result Achieves O ( T ) O(\sqrt{T}) O ( T ) regret bound for general weakly communicating MDPs. Logarithmic regret for continuous-time reinforcement learning.
problem Continuous-time Markov decision processes with unknown transition probabilities and holding times.
method Upper confidence reinforcement learning, mean holding time estimation, stochastic comparison of point processes.
result Logarithmic regret bound achieved in finite time.
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.
Adam optimization algorithm can have non-zero average regret under certain conditions.
problem Non-zero average regret in Adam optimization algorithm.
method Used a three-periodic sequence of linear functions on [-1,1] with slopes c, -1, -1, and analyzed Adam variants.
result Adam optimization algorithm can have non-zero average regret under certain conditions.
Actor-Critic method achieves optimal regret for unichain MDPs.
problem Scalable regret analysis for infinite-horizon average-reward MDPs.
method NAC-B, a Natural Actor-Critic with batching.
result Order-optimal regret of i l d e O ( T ) ilde{O}(\sqrt{T}) i l d e O ( T ) in infinite-horizon average-reward MDPs. The paper tackles lifelong learning in multi-armed bandits, aiming to minimize average regret over multiple tasks.
problem Minimizing average regret in multi-armed bandits over multiple tasks.
method Confidence interval tuning of UCB algorithms and greedy algorithms applied to a bandit over bandit approach.
result Empirical improvement over previous work in the mortal bandit problem.
We consider reinforcement learning (RL) in Markov Decision Processes in which an agent repeatedly interacts with an environment that is modeled by a controlled Markov process. At each time step t t t , it earns a reward, and also incurs a cost-vector consisting of M M M costs. We design model-based RL algorithms that maximi…
Adaptive designs achieve strong Neyman regret guarantees for ATE estimation.
problem Estimating unbiased average treatment effect in sequential experiments.
method Proposed adaptive designs with O ~ ( log T ) \widetilde{O}(\log T) O ( log T ) Neyman regret under boundedness assumptions and O ~ ( T ) \widetilde{O}(\sqrt{T}) O ( T ) multigroup Neyman regret in covariate-based settings. result Adaptive designs outperform non-adaptive designs in terms of Neyman regret, especially in covariate-based settings.
Kernel-based function approximation improves reinforcement learning performance.
problem Average reward reinforcement learning in infinite horizon settings.
method Optimistic algorithm based on kernel ridge regression.
result No-regret performance guarantees and confidence intervals for kernel-based predictions.
Paper proposes a model-free algorithm for CMDPs with long-term constraints, achieving optimal regret bounds.
problem Optimizing systems with long-term constraints where transition probabilities are unknown.
method Combines concepts from constrained optimization and Q-learning to propose an algorithm.
result Achieves optimal regret bounds for reward and constraint violation.
The paper tackles batch policy learning in Markov Decision Processes, focusing on average reward maximization.
problem Maximizing long-term average reward in Markov Decision Processes with batch learning.
method Doubly robust estimator for average reward, optimization algorithm for optimal policy, finite-sample regret guarantee.
result The proposed method achieves semiparametric efficiency and provides a finite-sample regret guarantee.
New algorithm optimizes Hölder continuous functions efficiently.
problem Optimizing Hölder continuous multivariate functions.
method Uses a query creation rule for global optimization, avoiding proxy functions.
result Achieves an average regret bound of $O(T^{- racα{n}})$ for Hölder exponent α α α . New method achieves small-loss regret bounds in random-order model.
problem Online learning with adversarial loss functions in random order.
method Extending batch-to-online transformation, using average sensitivity and stability.
result Small-loss regret bounds of order i l d e O ( φ ⋆ ( O P T T ) ) ilde O(\varphi^{\star}(\mathrm{OPT}_T)) i l d e O ( φ ⋆ ( OPT T )) . Model-free reinforcement learning is known to be memory and computation efficient and more amendable to large scale problems. In this paper, two model-free algorithms are introduced for learning infinite-horizon average-reward Markov Decision Processes (MDPs). The first algorithm reduces the problem to the discounted-r…
UCRL2-VTR achieves nearly optimal regret for learning MDPs with linear function approximation.
problem Learning infinite-horizon average-reward MDPs with linear function approximation.
method UCRL2-VTR algorithm with Bernstein-type bonus.
result Achieves a regret of i l d e O ( d D T ) ilde{O}(d\sqrt{DT}) i l d e O ( d D T ) with matching lower bound. New algorithm reduces regret in infinite MDPs with optimal variance-dependent bounds.
problem Infinite horizon MDPs lack optimal algorithms with low regret.
method Developed a UCB-style algorithm for average-reward and γ-regret.
result Achieved optimal variance-dependent regret bounds for both objectives.
New algorithms for learning MDPs with linear approximations in infinite-horizon settings.
problem Learning infinite-horizon average-reward MDPs with linear function approximation.
method Optimism principle, adversarial linear bandits, Natural Policy Gradient.
result Efficient algorithms with optimal or near-optimal regret bounds.
New algorithm reduces reinforcement learning regret to sqrt(T) without strong dynamics assumptions.
problem Infinite-horizon average-reward reinforcement learning with linear MDPs.
method Approximate by discounted-reward MDPs and apply optimistic value iteration.
result Achieves O(sqrt(T)) regret with polynomial complexity.
New algorithm for average reward learning with bounded hitting time assumption.
problem Minimizing regret in average reward reinforcement learning with bounded hitting time.
method Optimistic Q-learning with a novel L ‾ \overline{L} L operator for bounded hitting time. result Regret bound of i l d e O ( H 5 S A T ) ilde{O}(H^5 S\sqrt{AT}) i l d e O ( H 5 S A T ) for average reward learning. In this paper, we consider the problem of sequentially optimizing a black-box function f f f based on noisy samples and bandit feedback. We assume that f f f is smooth in the sense of having a bounded norm in some reproducing kernel Hilbert space (RKHS), yielding a commonly-considered non-Bayesian form of Gaussian process …
Adaptive clustering and personalization algorithms minimize regret in multi-agent stochastic linear bandits.
problem Minimizing regret in a multi-agent stochastic linear bandits framework with user heterogeneity.
method Proposes a novel algorithm that refines cluster identities and minimizes regret, adapting to cluster separation and user parameter deviations.
result Regret scales as O ( T / N ) \mathcal{O}(\sqrt{T/N}) O ( T / N ) for well-separated clusters and O ( T 1 2 + ε / ( N ) 1 2 − ε ) \mathcal{O}(T^{\frac{1}{2} + \varepsilon}/(N)^{\frac{1}{2} -\varepsilon}) O ( T 2 1 + ε / ( N ) 2 1 − ε ) for poorly separated clusters. The paper analyzes the sliding regret of stochastic bandit algorithms.
problem Measuring the one-shot behavior of no-regret algorithms in stochastic bandits.
method Introducing sliding regret to measure the worst pseudo-regret over a time-window.
result Randomized methods have optimal sliding regret, while index policies have the worst possible sliding regret.
New rule reduces exploration regret to logarithmic, improving bad episode handling.
problem Improving exploration regret in average reward MDPs.
method Replacing Doubling Trick with Vanishing Multiplicative rule in EVI-based algorithms.
result Regret is logarithmic under the new rule, significantly better than linear.
New geometric metrics improve Bayesian optimization performance evaluation.
problem Current metrics lack geometric insights and cannot compare algorithms effectively.
method Proposed four geometric metrics: precision, recall, average degree, and average distance.
result Proposed metrics provide more detailed evaluation of Bayesian optimization.
Most bandit algorithm designs are purely theoretical. Therefore, they have strong regret guarantees, but also are often too conservative in practice. In this work, we pioneer the idea of algorithm design by minimizing the empirical Bayes regret, the average regret over problem instances sampled from a known distributio…
Improved prediction algorithm for 'easy' sequences with reduced regret.
problem Prediction with expert advice for 'easy' sequences.
method Variant of NormalHedge algorithm using second-order ε ε ε -quantile regret bound. result Second-order ε ε ε -quantile regret bound of O ( V T log ( V T / ε ) ) O\big(\sqrt{V_T \log(V_T/ε)}\big) O ( V T log ( V T / ε ) ) for V T > log N V_T > \log N V T > log N . The paper derives a new theorem for predicting batches of data.
problem Finding lower bounds on minimal batch regret.
method Derives a conditional version of the regret-capacity theorem.
result Reveals a connection between conditional Rényi divergence and conditional Sibson's mutual information.
Unified meta-algorithm improves average performance across similar tasks in adversarial bandits.
problem Improving performance across multiple similar tasks in adversarial bandit settings.
method Unified meta-algorithm for multi-armed bandits and bandit linear optimization, tuning initialization, step-size, and entropy parameters.
result Unified meta-algorithm yields setting-specific guarantees for MAB and BLO, improving task-averaged regret.
Algorithm improves RL model selection for repeated games with utility maximization.
problem Optimal policy learning in repeated games with unknown opponent strategy.
method Proposes MRBEAR for average reward RL, applying to utility maximization in repeated games.
result Regret bound shows linear dependence on number of model classes in average reward RL.
The paper tackles learning to control systems with unknown parameters using Brownian noise.
problem Learning to control systems with unknown parameters.
method Proposes algorithms based on moving empirical averages and integrates statistical methods with stochastic control theory.
result Achieves a logarithmic expected regret rate.
OMD and DA perform similarly in static settings but OMD is inferior under dynamic learning rates.
problem Proving and understanding the performance difference between OMD and DA under dynamic learning rates.
method Introducing stabilization to OMD and modifying its convergence analysis.
result OMD with stabilization and DA have the same performance guarantees under dynamic learning rates.
A new algorithm avoids worst-case outcomes in risky contexts.
problem Risk-averse behavior in contextual bandits is challenging.
method Developed a first risk-averse contextual bandit algorithm with online regret guarantees.
result First algorithm with an online regret guarantee for risk-averse contextual bandits.
New algorithm LOOP learns infinite-horizon AMDPs efficiently with function approximation.
problem Learning optimal policies in infinite-horizon AMDPs with function approximation.
method LOOP combines model-based and value-based methods with novel confidence sets and policy updating.
result LOOP achieves sublinear regret bound of i l d e O ( p o l y ( d , s p ( V ∗ ) ) T β ) ilde{\mathcal{O}}(\mathrm{poly}(d, \mathrm{sp}(V^*)) \sqrt{Tβ} ) i l d e O ( poly ( d , sp ( V ∗ )) T β ) . Paper proposes ClipSMT algorithm for better ATE estimation.
problem Adaptive estimation of Average Treatment Effect (ATE).
method ClipSMT algorithm for improved Neyman regret.
result Achieves exponential improvements in Neyman regret.
Meta algorithm solves multivariate optimization using univariate optimizers.
problem Multivariate global optimization problems.
method Meta algorithm combining univariate global optimizers.
result Meta algorithm provides robust regret guarantees.
Meta-learning improves performance across similar tasks in adversarial bandit settings.
problem Improving performance across multiple similar tasks in adversarial bandit scenarios.
method Designing meta-algorithms that combine outer learners to tune hyperparameters of inner learners for MAB and BLO.
result Meta-algorithms improve task-averaged regret for MAB and BLO, showing direct relationship with action space-dependent measures.
Optimal algorithm found for collaborative learning in bandits with optimal regret bounds.
problem Minimizing regret in collaborative multi-agent bandit problems.
method Proposed an algorithm with optimal regret bounds for collaborative multi-agent multi-armed bandit model.
result First algorithm with order optimal regret bounds for collaborative bandit model.
PSRL extension for continuing environments reduces regret.
problem Formalizing and analyzing resampling approach for reinforcement learning.
method Continuing PSRL maintains a model of the environment and replaces it with samples from the posterior distribution.
result Established an i l d e O ( τ S A T ) ilde{O}(τS \sqrt{A T}) i l d e O ( τ S A T ) bound on Bayesian regret. New method evaluates LLMs fairness in universal prediction.
problem Evaluating fairness of large language models in universal prediction.
method Introducing batch regret as a modification of average regret for LLMs.
result Asymptotical value of batch regret for add-constant predictors on memoryless and first-order Markov sources.
Transforms offline algorithms to online with low regret in random order model.
problem Developing online algorithms with low approximate regret from offline approximation algorithms.
method General reduction theorem and coreset construction method.
result Achieves polylogarithmic ε-approximate regret for various online problems.
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 policy regret in tallying bandits.
problem Measuring online learning performance against adaptive adversaries.
method Tallying bandit model, efficient algorithm with complete policy regret guarantee.
result Achieves a complete policy regret guarantee of i l d e O ( m K T ) ilde{\mathcal{O}}(mK\sqrt{T}) i l d e O ( m K T ) . Paper presents an algorithm for optimal regret in communicating Markov decision processes.
problem Achieving optimal regret in Markov decision processes with a communicating assumption.
method The algorithm explicitly tracks the constant K(M) to learn optimally, balancing exploration, co-exploration, and exploitation.
result The algorithm achieves asymptotically optimal regret K ( M ) log ( T ) + o ( log ( T ) ) K(M) \log(T) + \mathrm{o}(\log(T)) K ( M ) log ( T ) + o ( log ( T )) for communicating Markov decision processes. Improved POMDP regret to sqrt(T) with known observation model.
problem Average-reward POMDPs with unknown transition model but known observation model.
method Optimistic algorithm using deterministic policies and novel estimation techniques.
result First approach with regret guarantee of sqrt(T) against optimal policy.
We study the problem of adaptive control of a high dimensional linear quadratic (LQ) system. Previous work established the asymptotic convergence to an optimal controller for various adaptive control schemes. More recently, for the average cost LQ problem, a regret bound of O ( T ) {O}(\sqrt{T}) O ( T ) was shown, apart form logarit…