We propose RUDDER, a novel reinforcement learning approach for delayed rewards in finite Markov decision processes (MDPs). In MDPs the Q-values are equal to the expected immediate reward plus the expected future rewards. The latter are related to bias problems in temporal difference (TD) learning and to high variance p…
On-device research index
arXiv research
A locally-built, LLM-digested index of recent arXiv papers in quant finance, geometry/topology, and statistical ML — keyword search served straight from SQLite on this machine.
168,694 papers · 148 categories
Trend · papers per month
48 results for “expected reward”
New RL formulation for maximizing maximum reward in molecule generation.
problem Traditional RL frameworks do not fit real-world applications like drug discovery.
method Formulated a new objective function to maximize maximum reward, derived Bellman equation, introduced operators, and proved convergence.
result Achieved state-of-the-art results in molecule generation.
New approach to multi-armed bandit problem aims to maximize highest total reward.
problem Traditional multi-armed bandit problem objective of maximizing total reward is not suitable in certain applications.
method Adaptive explore-then-commit policy with confidence bounds and adaptive stopping criterion.
result Achieves asymptotic and worst-case regret bounds for the new objective.
Framework for optimizing search engine rankings using observational data.
problem Optimizing ranking policies for search engines using limited observational data.
method Formulated expected reward optimization problem, estimated context value distribution, trained ranking policy via Bayesian inference.
result Demonstrated trade-offs in ranking policies trained on empirical reward estimates.
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.
We propose a new complexity measure for Markov decision processes (MDPs), the maximum expected hitting cost (MEHC). This measure tightens the closely related notion of diameter [JOA10] by accounting for the reward structure. We show that this parameter replaces diameter in the upper bound on the optimal value span of a…
New concept: reward hacking, where optimizing a flawed reward function can hurt performance.
problem Optimizing imperfect reward functions leads to poor performance.
method Formal definition and analysis of reward hacking, examining conditions for unhackability.
result Reward functions are usually hackable, making it hard to align AI with human values.
Study of repeated games with unobserved agent rewards using MAB framework.
problem Designing policies for principals in repeated principal-agent games with unobservable agent rewards.
method Developed a policy achieving low regret (square-root regret up to a log factor) for perfect-knowledge agents.
result Constructed an estimator for agent's expected reward and designed a policy achieving low regret.
Vanishing gradients hinder reinforcement finetuning of language models.
problem Vanishing gradients impede the optimization of language models using reinforcement finetuning.
method The study identifies vanishing gradients as a fundamental optimization obstacle in reinforcement finetuning and proposes an initial supervised finetuning phase to mitigate this issue.
result An initial supervised finetuning phase is crucial for successful reinforcement finetuning of language models, as it helps prevent vanishing gradients and maximizes rewards.
We consider a multiobjective multiarmed bandit problem with lexicographically ordered objectives. In this problem, the goal of the learner is to select arms that are lexicographic optimal as much as possible without knowing the arm reward distributions beforehand. We capture this goal by defining a multidimensional for…
The Perils of Optimizing Learned Reward Functions: Low Training Error Does Not Guarantee Low Regretcs.LG
Learning reward functions can lead to poor policy performance despite low error.
problem Low error in learned reward functions does not guarantee low regret in policy performance.
method Mathematical analysis of reward learning and policy optimization.
result A low expected test error of the reward model guarantees low worst-case regret, but error-regret mismatch can occur with certain data distributions.
Investigates risk measures for DC pension decumulation.
problem Develop optimal decumulation strategies for DC plan holders.
method Formulates decumulation as a control problem, studies risk measures (expected shortfall, linear shortfall, probability of shortfall).
result Optimal controls for expected reward and expected shortfall are identical to those for expected reward and linear shortfall.
A new game-theoretic approach balances downside risk with expected reward.
problem Traditional game theory views risk only from the upside perspective, ignoring downside risk.
method Introduces downside risk aware equilibria (DRAE) based on lower partial moments.
result Successfully finds equilibria that balance downside risk with expected reward.
New algorithms ensure fair selection in combinatorial semi-bandit with unrestricted delays.
problem Fair selection in stochastic combinatorial semi-bandit with delayed feedback.
method Introduced merit-based fairness constraints and new bandit algorithms for reward and fairness.
result Achieved sublinear expected reward and fairness regrets with dependence on delay distribution quantiles.
Rotting Banditsstat.ML
The Multi-Armed Bandits (MAB) framework highlights the tension between acquiring new knowledge (Exploration) and leveraging available knowledge (Exploitation). In the classical MAB problem, a decision maker must choose an arm at each time step, upon which she receives a reward. The decision maker's objective is to maxi…
We develop a theory for solving continuous time optimal stopping problems for non-linear expectations. Our motivation is to consider problems in which the stopper uses risk measures to evaluate future rewards.
BAICS identifies best arm with fairness constraints on subpopulations.
problem Identify the best arm while ensuring fairness across subpopulations.
method Formulated and solved BAICS problem, analyzed complexity, designed algorithm.
result Algorithm's sample complexity matches theoretical lower bound.
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…
We study, to the best of our knowledge, the first Bayesian algorithm for unimodal Multi-Armed Bandit (MAB) problems with graph structure. In this setting, each arm corresponds to a node of a graph and each edge provides a relationship, unknown to the learner, between two nodes in terms of expected reward. Furthermore, …
Algorithm maximizes rewards with a budget and giving up option.
problem Sequential decision-making with stochastic rewards and resource consumption.
method Upper Confidence Bound (UCB) algorithm for maximizing cumulative reward.
result Logarithmic regret bound with improved dependence on problem parameters.
This work extends reinforcement learning to handle non-cumulative objectives.
problem Optimizing functions of rewards rather than their sum in decision processes.
method Mapping NCMDPs to standard MDPs for reinforcement learning.
result Reinforcement learning techniques can be applied to NCMDPs.
One obstacle to applying reinforcement learning algorithms to real-world problems is the lack of suitable reward functions. Designing such reward functions is difficult in part because the user only has an implicit understanding of the task objective. This gives rise to the agent alignment problem: how do we create age…
SIBRE boosts reinforcement learning convergence by rewarding improvement over past performance.
problem Improving the rate of convergence in reinforcement learning.
method SIBRE is a reward shaping approach that rewards improvement over the agent's own past performance.
result SIBRE converges faster and more stably to the optimal policy compared to baseline RL algorithms.
Classical multi-armed bandit problems use the expected value of an arm as a metric to evaluate its goodness. However, the expected value is a risk-neutral metric. In many applications like finance, one is interested in balancing the expected return of an arm (or portfolio) with the risk associated with that return. In …
New method for LLMs to learn reasoning by optimizing latent variables.
problem Teaching LLMs to generate logical justifications for answers.
method Formalized reasoning as latent variable model, derived FEM objective, designed sampling schemes.
result Prompt Posterior Sampling (PPS) outperforms other schemes in learning to reason.
The paper introduces a new intrinsic reward method for exploration in reinforcement learning.
problem Improving exploration in reinforcement learning agents.
method Intrinsic rewards proportional to the entropy of future state-action features.
result The new objective leads to improved visitation of features within individual trajectories.
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.
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.
f-GRPO and Beyond: Divergence-Based Reinforcement Learning Algorithms for General LLM Alignmentcs.LG
Extends reinforcement learning alignment to scalar rewards, improving math reasoning.
problem Designing reinforcement learning algorithms for general LLM alignment.
method Introduces f-GRPO and f-HAL, estimating f-divergences between reward-aligned and unaligned distributions.
result Improves math reasoning RLVR tasks and mitigates reward hacking.
In reinforcement learning the Q-values summarize the expected future rewards that the agent will attain. However, they cannot capture the epistemic uncertainty about those rewards. In this work we derive a new Bellman operator with associated fixed point we call the `knowledge values'. These K-values compress both the …
New algorithm identifies good arms with fewer samples when thresholds are close.
problem Good arm identification in bandit problems with small threshold gaps.
method Proposes lil'HDoC algorithm to improve GAI under small threshold gaps.
result Sample complexity of first λ output arm is nearly identical to HDoC algorithm when thresholds are close.
In this paper we propose a novel framework for decentralized, online learning by many learners. At each moment of time, an instance characterized by a certain context may arrive to each learner; based on the context, the learner can select one of its own actions (which gives a reward and provides information) or reques…
We propose a bandit algorithm that explores by randomizing its history of rewards. Specifically, it pulls the arm with the highest mean reward in a non-parametric bootstrap sample of its history with pseudo rewards. We design the pseudo rewards such that the bootstrap mean is optimistic with a sufficiently high probabi…
A new reward shaping method balances learning efficiency and effectiveness for robot manipulation.
problem Efficient and effective learning in robot manipulations with system uncertainty.
method Dense2Sparse reward shaping method combining dense and sparse rewards.
result Dense2Sparse method achieves higher expected reward and better system uncertainty tolerance.
The goal of the inverse reinforcement learning (IRL) problem is to recover the reward functions from expert demonstrations. However, the IRL problem like any ill-posed inverse problem suffers the congenital defect that the policy may be optimal for many reward functions, and expert demonstrations may be optimal for man…
There exist a number of reinforcement learning algorithms which learnby climbing the gradient of expected reward. Their long-runconvergence has been proved, even in partially observableenvironments with non-deterministic actions, and without the need fora system model. However, the variance of the gradient estimator ha…
Novel alignment framework for text-to-image generation using diffusion models and flow matching.
problem Improving text-to-image generation with minimal computational resources.
method Proposes a novel alignment framework that decomposes the score function into pre-trained score plus a conditional expectation of the reward.
result Achieves comparable performance to finetuning-based models with reduced computational cost.
The most data-efficient algorithms for reinforcement learning in robotics are model-based policy search algorithms, which alternate between learning a dynamical model of the robot and optimizing a policy to maximize the expected return given the model and its uncertainties. However, the current algorithms lack an effec…
Optimal algorithm for maximizing rewards in contextual bandits with resource constraints.
problem Maximizing rewards in contextual bandits with resource constraints.
method Proposed a universal and optimal algorithmic framework for CBwK by reducing it to online regression.
result Established the optimality of the proposed algorithm for various function classes.
Text generation is a crucial task in NLP. Recently, several adversarial generative models have been proposed to improve the exposure bias problem in text generation. Though these models gain great success, they still suffer from the problems of reward sparsity and mode collapse. In order to address these two problems, …
New policy for non-stationary multi-armed bandits with recovering rewards.
problem Non-stationary multi-armed bandits with recovering rewards.
method Purely Periodic Policies integrating upper confidence bound procedure.
result Approximately regret against the offline benchmark.
Experimentally, it has been observed that humans and animals often make decisions that do not maximize their expected utility, but rather choose outcomes randomly, with probability proportional to expected utility. Probability matching, as this strategy is called, is equivalent to maximum entropy reinforcement learning…
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.
Optimizing rewards under budget constraints with correlated costs and rewards.
problem Maximizing total expected reward under a budget constraint on total cost with correlated and potentially heavy-tailed cost-reward pairs.
method Proposes algorithms exploiting correlation between cost and reward via linear minimum mean-square error estimation to achieve tight regret bounds.
result Achieves regret for a budget under certain moment conditions.
Learning robust value functions given raw observations and rewards is now possible with model-free and model-based deep reinforcement learning algorithms. There is a third alternative, called Successor Representations (SR), which decomposes the value function into two components -- a reward predictor and a successor ma…
Paper establishes DRL for high-dimensional rewards.
problem Intractable reinforcement learning with high-dimensional rewards.
method Theoretical foundations and a novel DRL algorithm.
result Bellman operator contraction in high-dimensional spaces.
We study contextual bandits with budget and time constraints, referred to as constrained contextual bandits.The time and budget constraints significantly complicate the exploration and exploitation tradeoff because they introduce complex coupling among contexts over time.Such coupling effects make it difficult to obtai…
New algorithm for multi-player bandits without needing lower bounds or scaling inversely.
problem Multi-player bandits without collision sensing information.
method Proposes a novel algorithm that circumvents two problems of existing algorithms.
result Proves a theoretical regret upper bound and shows superior performance in practice.