New RL method MAC improves performance in sparse reward settings.
problem Slow mixing in large state spaces or sparse rewards.
method Multi-level Monte Carlo Actor-Critic (MAC) algorithm.
result Achieves convergence rate comparable to state-of-the-art AC algorithms.
New algorithm learns optimal policy for average reward MDPs with sample complexity matching lower bound.
problem Learning optimal policy for average reward in uniformly ergodic MDPs.
method Developed an estimator with sample complexity of O(|S||A|t_{mix}ε^{-2}).
result First algorithm to match lower bound of existing literature.
Algorithm learns near-optimal policies for reward-mixing MDPs with few latent contexts.
problem Episodic reinforcement learning in reward-mixing Markov decision processes with a few latent contexts.
method Sample-efficient algorithm EM^2 using higher-order method-of-moments approach.
result Provides an ε-optimal policy using O(ε^(-2) * S^d A^d * poly(H, Z)^d) episodes for arbitrary M ≥ 2.
Study improves reinforcement learning for stable long-term performance.
problem Distributionally robust average-reward reinforcement learning for stable long-term performance.
method Proposes two algorithms to achieve near-optimal sample complexity.
result Achieves a sample complexity of O ( ∣ S ∣ ∣ A ∣ t m i x 2 ε − 2 ) O(|\mathbf{S}||\mathbf{A}| t_{\mathrm{mix}}^2\varepsilon^{-2}) O ( ∣ S ∣∣ A ∣ t mix 2 ε − 2 ) for estimating optimal policy and robust average reward. New algorithm for collaborative bandit learning reduces sample complexity and regret.
problem Optimal arm identification in a multi-agent bandit model with communication.
method Phased elimination with data-dependent sampling schemes.
result Near-optimal algorithm for pure exploration in collaborative bandit learning.
RoME optimizes mobile health interventions by modeling user and time-specific effects.
problem Challenges in optimizing mobile health interventions due to participant heterogeneity, nonstationarity, and nonlinear relationships.
method RoME uses a Robust Mixed-Effects contextual bandit algorithm with random effects, network cohesion penalties, and debiased machine learning.
result RoME achieves robust regret bounds even with complex baseline rewards, demonstrating superior performance in simulations and studies.
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.
We derive and analyze learning algorithms for apprenticeship learning, policy evaluation, and policy gradient for average reward criteria. Existing algorithms explicitly require an upper bound on the mixing time. In contrast, we build on ideas from Markov chain theory and derive sampling algorithms that do not require …
Paper tackles best mixed arm identification with cost constraints in bandit models.
problem Finding the best mixed arm with cost constraints in a stochastic bandit model.
method Proposes SFSR algorithm combining successive reject and score-function-based rejection criteria.
result Upper and lower bounds on mis-identification probability show exponential decay with budget.
Optimal sample complexity analysis for plug-in approach in average-reward MDPs.
problem Learning optimal policies in average-reward MDPs with a generative model.
method Plug-in approach that constructs a model estimate and computes an optimal policy.
result Optimal sample complexities for the plug-in approach without prior knowledge of problem parameters.
The paper tackles reinforcement learning with exogenous variables and rewards.
problem Exogenous state variables and rewards slow reinforcement learning by introducing uncontrolled variation.
method Formalizes exogenous state variables and rewards, decomposes MDP into exogenous and endogenous components, and introduces algorithms to discover these components.
result Optimal policies for the endogenous MDP are also optimal for the original MDP, but the endogenous MDP is easier to solve due to reduced variance.
New RL theory reduces sample complexity for mixing MDPs.
problem Optimal sample complexity for reinforcement learning in mixing MDPs.
method Regeneration-type ideas to analyze mixing times.
result Optimal sample complexity depends on mixing time, not just discount factor.
Optimizes learning policies in average-reward MDPs with improved sample complexity.
problem Learning optimal policies in average-reward MDPs with limited samples.
method Reduces to discounted MDPs and uses improved bounds for variance parameters.
result Establishes minimax optimal sample complexity bound of O(SA(H/ε^2))
Study non-parametric value function estimation from a single path.
problem Estimating value function from a single trajectory in Markov reward processes.
method Kernel-based multi-step temporal difference (TD) estimates, including K K K -step look-ahead TD and TD ( λ ) (λ) ( λ ) . result Non-asymptotic guarantees for TD estimates, capturing interactions between mixing time and model mis-specification.
ORIL learns a reward function from unlabeled data to improve robot learning.
problem Leveraging unlabeled data for robot learning.
method ORIL learns a reward function from demonstrator and unlabeled trajectories, annotates data, and trains an agent via offline reinforcement learning.
result ORIL consistently outperforms BC agents on various robotic tasks.
New algorithm balances user reward and statistical inference by mixing TS with UR based on difference size.
problem Combining statistical inference with user reward in adaptive experiments.
method TS-PostDiff algorithm that uses UR when differences are small and TS when large.
result TS-PostDiff reduces false positives and increases statistical power for small differences, while maximizing reward for large ones.
Paper explores limits and possibilities of aligning LLMs with human preferences.
problem Aligning LLMs with diverse human preferences to ensure fairness and informed outcomes.
method Analysis of probabilistic representation of human preferences and preservation of diverse preferences.
result LLMs can't fully align with human preferences using reward-based approaches due to Condorcet cycles, but mixed strategies are statistically possible.
Theory for RLHF generalization under reward shift and clipped KL.
problem Theoretical understanding of RLHF generalization, especially with reward shift and clipped KL.
method Developed generalization theory for RLHF, accounting for reward shift and clipped KL.
result Presented generalization bounds for RLHF, suggesting generalization error from sampling, reward shift, and KL clipping.
Proposes meTS for efficient exploration in correlated bandits.
problem Efficiently explore in bandits with correlated actions.
method Mixed-effect model with Thompson Sampling.
result Bound on Bayes regret with two terms.
Autonomous cyber-physical agents and systems play an increasingly large role in our lives. To ensure that agents behave in ways aligned with the values of the societies in which they operate, we must develop techniques that allow these agents to not only maximize their reward in an environment, but also to learn and fo…
New framework shifts bandit algorithms from expected reward to preference metrics, optimizing mixtures of arms.
problem Traditional bandit algorithms focus on expected rewards, ignoring variability and risk.
method Introduces preference metrics (PMs) and designs algorithms to optimize mixtures of arms.
result Optimal policy selects mixtures of arms based on specific weights, not a single best arm.
Efficient dispatching rule in manufacturing industry is key to ensure product on-time delivery and minimum past-due and inventory cost. Manufacturing, especially in the developed world, is moving towards on-demand manufacturing meaning a high mix, low volume product mix. This requires efficient dispatching that can wor…
We consider Markov Decision Processes (MDPs) where the rewards are unknown and may change in an adversarial manner. We provide an algorithm that achieves state-of-the-art regret bound of O ( τ ( ln ∣ S ∣ + ln ∣ A ∣ ) T ln ( T ) ) O( \sqrt{τ(\ln|S|+\ln|A|)T}\ln(T)) O ( τ ( ln ∣ S ∣ + ln ∣ A ∣ ) T ln ( T )) , where S S S is the state space, A A A is the action space, τ τ τ is the mixing time of the MDP, and $…
We study a strategic version of the multi-armed bandit problem, where each arm is an individual strategic agent and we, the principal, pull one arm each round. When pulled, the arm receives some private reward v a v_a v a and can choose an amount x a x_a x a to pass on to the principal (keeping v a − x a v_a-x_a v a − x a for itself). All non-pulle…
What is a good exploration strategy for an agent that interacts with an environment in the absence of external rewards? Ideally, we would like to get a policy driving towards a uniform state-action visitation (highly exploring) in a minimum number of steps (fast mixing), in order to ease efficient learning of any goal-…
New offline RL method handles average-reward MDPs with single-policy coverage.
problem Challenges in offline reinforcement learning due to distribution shift and non-uniform coverage.
method Develops an algorithm based on pessimistic discounted value iteration with quantile clipping.
result First fully single-policy sample complexity bound for average-reward offline RL.
QR-MIX models joint state-action values as a distribution to handle randomness in MARL.
problem Randomness in rewards and observations leads to randomness in long-term returns in MARL.
method QR-MIX uses quantile regression and combines it with QMIX and IQN to model joint state-action values as a distribution.
result QR-MIX outperforms QMIX in the StarCraft Multi-Agent Challenge (SMAC) environment.
In reinforcement learning (RL) , one of the key components is policy evaluation, which aims to estimate the value function (i.e., expected long-term accumulated reward) of a policy. With a good policy evaluation method, the RL algorithms will estimate the value function more accurately and find a better policy. When th…
A novel algorithm minimizes regret in a multi-agent bandit problem with time-varying random graphs and heterogeneous rewards.
problem Minimizing regret in a multi-agent multi-armed bandit problem with time-varying random graphs and heterogeneous rewards.
method Introduces a novel algorithmic framework combining averaging-based consensus with a weighting technique and upper confidence bound.
result Derives optimal instance-dependent regret upper bounds of order log T \log{T} log T in both sub-gaussian and sub-exponential environments. RLVR training dynamics reveal an implicit curriculum that shapes learning progression.
problem Understanding how RLVR overcomes the long-horizon barrier.
method Developed a theory of training dynamics for RLVR on transformers, using Fourier analysis on finite groups.
result Mixed-difficulty training naturally follows an implicit curriculum, shaping the learning progression from easy to hard.
Algorithm finds near-optimal VaR portfolios using MILP, improving risk management.
problem Computing optimal VaR portfolios is hard due to non-convexity and combinatorial nature.
method Formulates VaR portfolio problem as MILP, uses alternate formulations for guarantees.
result Near-optimal VaR portfolios with near-optimality guarantees.
DSAC improves cooperative MARL with general utilities, converging faster than existing methods.
problem Improving cooperation in multi-agent reinforcement learning with nonlinear utilities.
method Decentralized Shadow Reward Actor-Critic (DSAC) that estimates local occupancy measures and derivatives.
result DSAC converges to ε-stationarity in O(1/ε^2.5) steps with high probability, finding globally optimal policies.
Our goal is for agents to optimize the right reward function, despite how difficult it is for us to specify what that is. Inverse Reinforcement Learning (IRL) enables us to infer reward functions from demonstrations, but it usually assumes that the expert is noisily optimal. Real people, on the other hand, often have s…
Unified framework for solving MDPs with stochastic mirror descent.
problem Approximately solving infinite-horizon Markov decision processes (MDPs).
method Primal-dual stochastic mirror descent for MDPs with a unified framework.
result Computes ε-optimal policies with expected samples for both average-reward and discounted MDPs.
Algorithm for low-rank matrix bandits with heavy-tailed rewards, achieving nearly optimal regret bound.
problem Stochastic low-rank matrix bandit with heavy-tailed rewards.
method LOTUS algorithm using truncation and dynamic exploration.
result Regret bound of order $ ilde O(d^ rac{3}{2}r^ rac{1}{2}T^ rac{1}{1+δ}/ ilde{D}_{rr})$ without knowing T T T . Optimizes learning policies in MDPs with weakly communicating structure.
problem Learning optimal policies in weakly communicating MDPs with generative model.
method Span-based approach, reducing to discounted MDPs for analysis.
result First minimax optimal sample complexity bound for weakly communicating MDPs.
New algorithm for optimizing statistical utilities in bandits.
problem Optimizing statistical functionals of long-run reward distributions.
method Influence-function calculus for stochastic gradient estimation, entropic mirror-ascent algorithm.
result Regret bounds that separate optimization and estimation errors.
Study uses RL to optimize global equity portfolios, finds mixed results.
problem Optimizing dynamic portfolio weights across diverse global markets.
method Deep reinforcement learning with Soft Actor-Critic, incorporating various constraints and reward formulations.
result RL strategies achieve competitive performance, but no strategy consistently outperforms Buy and Hold.
New algorithm achieves nearly optimal regret with one-pass updates for GLB problems.
problem Generalized linear bandits with non-linear reward distributions.
method Jointly efficient algorithm using OMD estimator with one-pass updates.
result Nearly optimal regret bound with O ( 1 ) \mathcal{O}(1) O ( 1 ) time and space complexities per round. RLVR learning dynamics naturally create an implicit curriculum for transformers.
problem How rewards based on final outcomes help overcome the long-horizon barrier in reasoning models.
method Developed a theory of training dynamics for RLVR on transformers, using Fourier analysis on finite groups.
result Mixed-difficulty training induces an implicit curriculum that shapes the learning progression from easy to hard.
This work develops confidence intervals for off-policy evaluation.
problem Estimating expected reward with uncertainty quantification.
method Primal-dual optimization with kernel Bellman loss and martingale concentration inequality.
result Developed practical algorithm for non-asymptotic confidence intervals.
Reward hacking exploits misspecified rewards, affecting agent capabilities and true performance.
problem Reward hacking in RL models exploiting reward misspecifications.
method Constructed four RL environments with misspecified rewards; analyzed agent capabilities and behavior.
result More capable agents exploit reward misspecifications, achieving higher proxy reward but lower true reward.
Paper introduces PRMs to learn non-Markovian stochastic rewards for reinforcement learning.
problem Lack of structured representation for non-Markovian stochastic rewards in reinforcement learning.
method Introduces probabilistic reward machines (PRMs) and presents an algorithm to learn them from decision processes.
result Algorithm proves correct and convergent for learning PRMs from decision processes.
This work analyzes the value of future reward information in RL.
problem Analyzing the impact of knowing future rewards in reinforcement learning.
method Competitive analysis and worst-case reward distribution.
result Exact ratios between standard RL agents and those with future-reward lookahead.
Self-supervised reward prediction improves RL in sparse reward settings.
problem Data efficiency and sparse reward signals in reinforcement learning.
method Learning a state representation for reward prediction and using it to shape rewards.
result Self-supervised reward prediction enhances RL algorithms in single-goal environments.
The study categorizes reward errors in reinforcement learning, finding some can be beneficial.
problem Training language models with imperfect proxy rewards.
method Theoretical analysis of policy gradient optimization and categorization of reward errors.
result Reward errors can be benign or even beneficial, preventing policy from stalling.
Reward collapse occurs when ranking-based reward models yield uniform rewards for different prompts.
problem Reward collapse in aligning large language models with human preferences.
method Introduced a prompt-aware optimization scheme to derive closed-form expressions for reward distributions.
result Our prompt-aware utility functions significantly alleviate reward collapse during training.
Reward models need more than just accuracy for effective RLHF.
problem The effectiveness of reward models in RLHF is not fully understood.
method An optimization perspective to evaluate reward models.
result Reward models with low reward variance can lead to a flat optimization landscape, hindering performance.