New lower bounds for combinatorial multi-armed bandits for general reward functions.
problem Maximizing reward in sequential decisions with sets of arms.
method Proved tight regret lower bounds for all smooth reward functions under mild assumptions.
result Lower bounds are tight up to log-factors for monotone reward functions.
IDS improves RLHF by smoothing reward data, enhancing model performance.
problem Reward model performance degrades and overoptimization hinders true objective.
method Iterative Data Smoothing (IDS) updates model and data labels during each epoch.
result IDS outperforms traditional methods in RLHF.
Improved inference-time alignment using Best-of-N and smoothing.
problem Reward overoptimization in Best-of-N (BoN) due to poor proxy reward models.
method Introduced Soft Best-of-N (SBoN) and analyzed its performance through KL divergence and regret analysis.
result Smoothing helps SBoN mitigate reward overoptimization, especially when proxy reward quality is low.
This paper proposes a new algorithm for learning guidance rewards in RL.
problem Long-term temporal credit assignment in sparse or delayed reward environments.
method Surrogate RL objective with trajectory-space smoothing to learn guidance rewards.
result Guidance rewards can be learned without additional neural networks and have intuitive interpretation.
We consider the combinatorial multi-armed bandit (CMAB) problem, where the reward function is nonlinear. In this setting, the agent chooses a batch of arms on each round and receives feedback from each arm of the batch. The reward that the agent aims to maximize is a function of the selected arms and their expectations…
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.
In this paper we consider the problem of online stochastic optimization of a locally smooth function under bandit feedback. We introduce the high-confidence tree (HCT) algorithm, a novel any-time X-armed bandit algorithm, and derive regret bounds matching the performance of existing state-of-the-art in term…
We consider a discounted reward control problem in continuous time stochastic environment where the discount rate might be an unbounded function of the control process. We provide a set of general assumptions to ensure that there exists a smooth classical solution to the corresponding HJB equation. Moreover, some verif…
Imitation Learning describes the problem of recovering an expert policy from demonstrations. While inverse reinforcement learning approaches are known to be very sample-efficient in terms of expert demonstrations, they usually require problem-dependent reward functions or a (task-)specific reward-function regularizatio…
This paper optimizes attacks on stochastic bandits and proposes defenses against them.
problem Optimizing adversarial attacks on stochastic bandit algorithms.
method Designs optimal attack strategies and proposes defense algorithms.
result Optimal attack strategies and defense algorithms achieve near perfect performance.
Consider a nonparametric contextual multi-arm bandit problem where each arm a∈[K] is associated to a nonparametric reward function fa:[0,1]→R mapping from contexts to the expected reward. Suppose that there is a large set of arms, yet there is a simple but unknown structure amongst the arm reward…
A common approach for defining a reward function for Multi-objective Reinforcement Learning (MORL) problems is the weighted sum of the multiple objectives. The weights are then treated as design parameters dependent on the expertise (and preference) of the person performing the learning, with the typical result that a …
BERT embeddings improve sequence quality metrics.
problem Measuring the quality of generated sequences against references.
method Employ contextual BERT embeddings for sequence-level reward.
result Contextual embeddings provide a more effective learning signal.
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.
Optimal strategy proposed for maximizing cumulative reward in continuum-armed bandits.
problem Maximizing cumulative reward in a scenario with limited resources and unknown stochastic rewards.
method Proposed an optimal strategy for a nonparametric setting with side information on actions.
result Optimal regret scales as \(O(T^{1/3})\) up to poly-logarithmic factors when \(T\) is proportional to \(N\).
Adaptive smooth non-stationary bandits achieve optimal regret rates without knowing parameters.
problem Smooth non-stationary bandits with Hölder class rewards.
method Established optimal dynamic regret rate and adaptive algorithm.
result Optimal dynamic regret can be attained adaptively without knowing Hölder exponent and coefficient.
New algorithm optimizes smooth functions with Hölder exponent > 1.
problem Optimizing smooth functions with unknown Hölder exponent > 1.
method Two-layer algorithms using misspecified linear/polynomial bandit algorithms in bins.
result Regret bound of O~(Td+2αd+α) for α>1. In structured output prediction tasks, labeling ground-truth training output is often expensive. However, for many tasks, even when the true output is unknown, we can evaluate predictions using a scalar reward function, which may be easily assembled from human knowledge or non-differentiable pipelines. But searching th…
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…
Algorithm tackles adaptive discretization in adversarial Lipschitz bandits for dynamic pricing and auctions.
problem Adaptive discretization in adversarial Lipschitz bandits.
method Adversarial Zooming algorithm for adaptive discretization.
result First algorithm for adversarial Lipschitz bandits with instance-dependent regret bounds.
Develops statistical framework for resolving reward function ambiguity in inverse reinforcement learning.
problem Non-uniqueness of reward functions in inverse reinforcement learning.
method Entropy regularization combined with least-squares reconstruction of the reward from the soft Bellman residual.
result Least-squares reward function is unique and consistent with the expert policy.
We study the out-of-sample properties of robust empirical optimization problems with smooth φ-divergence penalties and smooth concave objective functions, and develop a theory for data-driven calibration of the non-negative "robustness parameter" δ that controls the size of the deviations from the nominal model. Bu…
A new ES method improves reinforcement learning speed and accuracy.
problem Slow convergence and local maxima in reinforcement learning.
method Directional Gaussian Smoothing Evolution Strategy (DGS-ES)
result DGS-ES accelerates RL training with high accuracy and nonlocal search direction.
In X-armed bandit problem an agent sequentially interacts with environment which yields a reward based on the vector input the agent provides. The agent's goal is to maximise the sum of these rewards across some number of time steps. The problem and its variations have been a subject of numerous studies, su…
New method extends low-rank MDPs to continuous action spaces.
problem Limited applicability of current low-rank MDP methods to continuous action spaces.
method Extending FLAMBE algorithm to continuous action spaces with Hölder smoothness conditions.
result Similar PAC bound achieved for continuous actions with polynomial dependence on smoothness order.
SPARKLE handles high-dimensional covariates for online decision-making.
problem Complex reward-covariate relationships in high-dimensional settings.
method SPARKLE uses a sparse additive reward model with doubly penalized estimator and adaptive screening.
result SPARKLE achieves sublinear regret bound logarithmic in covariate dimensionality.
Kernel-UCBVI algorithm balances exploration and exploitation in metric state-action spaces.
problem Exploration-exploitation dilemma in finite-horizon reinforcement learning with metric state-action spaces.
method Kernel-UCBVI, leveraging smoothness and kernel estimators of rewards and transitions.
result First regret bound for kernel-based RL using smoothing kernels, O(H3K2d/(2d+1)). Oriental ink painting, called Sumi-e, is one of the most appealing painting styles that has attracted artists around the world. Major challenges in computer-based Sumi-e simulation are to abstract complex scene information and draw smooth and natural brush strokes. To automatically find such strokes, we propose to mode…
In classical Q-learning, the objective is to maximize the sum of discounted rewards through iteratively using the Bellman equation as an update, in an attempt to estimate the action value function of the optimal policy. Conventionally, the loss function is defined as the temporal difference between the action value and…
DS-TS adapts to abrupt and smooth changes in bandit problems.
problem Non-stationary multi-armed bandit problems with abrupt and smooth changes.
method Discounted Thompson Sampling with Gaussian priors.
result Achieves nearly optimal regret bound for both abrupt and smooth changes.
Reinforcement learning requires manual specification of a reward function to learn a task. While in principle this reward function only needs to specify the task goal, in practice reinforcement learning can be very time-consuming or even infeasible unless the reward function is shaped so as to provide a smooth gradient…
SPReD uses uncertainty to decide imitation from demonstrations.
problem Learning from sparse rewards with few demonstrations.
method Ensemble methods to model Q-value distributions, probabilistic and advantage-based uncertainty quantification.
result Significant gains in reinforcement learning across multiple tasks.
This paper improves sample efficiency for off-policy evaluation with preference data.
problem Improving sample efficiency for off-policy evaluation with preference data.
method Using a deep neural network to learn the value function and leveraging manifold structure.
result Established a provably efficient guarantee for off-policy evaluation with RLHF.
TT-DAC-PS: A deterministic actor-critic approach for optimal trade execution
problem Optimal execution of large stock sell programs
method Twin-Target Deterministic Actor-Critic with Policy Smoothing
result Reduces mean implementation shortfall percentage
Unified framework for multi-user bandits using Laplacian kernels.
problem Multi-user contextual bandits with graph-related users and non-linear rewards.
method Joint penalty combining graph smoothness and individual roughness in a unified RKHS.
result Unified multi-user RKHS and effective dimension for regret bounds.
We introduce a simple extension of the minority game in which the market rewards contrarian (resp. trend-following) strategies when it is far from (resp. close to) efficiency. The model displays a smooth crossover from a regime where contrarians dominate to one where trend-followers dominate. In the intermediate phase,…
AOT aligns LLMs on distributional preferences via optimal transport.
problem Current LLM alignment techniques lack distributional level alignment.
method Alignment via Optimal Transport (AOT) aligns LLMs on unpaired preference data.
result AOT enables alignment by penalizing reward distribution violations.
New method uses LLMs to generate detailed scientific hypotheses.
problem Generating detailed, actionable scientific hypotheses from coarse initial directions.
method Hierarchical search method that incrementally adds details to hypotheses.
result Hierarchical search method consistently outperforms strong baselines on expert-annotated hypotheses.
OE2D framework reduces contextual bandits to offline regression for near-optimal regret.
problem Efficiently learning contextual bandits with large action spaces and complex reward functions.
method Offline Estimation to Decisions (OE2D) algorithm that reduces contextual bandits to offline regression.
result Near-optimal regret for contextual bandits with large action spaces and O(logT) calls to an offline regression oracle. Reward augmented maximum likelihood (RAML), a simple and effective learning framework to directly optimize towards the reward function in structured prediction tasks, has led to a number of impressive empirical successes. RAML incorporates task-specific reward by performing maximum-likelihood updates on candidate outpu…
A model used for velocity control during car following was proposed based on deep reinforcement learning (RL). To fulfil the multi-objectives of car following, a reward function reflecting driving safety, efficiency, and comfort was constructed. With the reward function, the RL agent learns to control vehicle speed in …
Investigates sequential problems on graph structures and large action spaces.
problem Sequential decision-making on graph structures and large action spaces.
method Spectral bandits, side observations, influence maximization, kernel bandits, polymatroid bandits, function optimization, infinitely many-arms bandits.
result Contributions to graph and structured bandits.
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.
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.
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.
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.