Study evaluates new models using human feedback from another model.
problem Evaluate a new model using human feedback collected for another model.
method Formalize problem, propose model-based and model-free estimators, analyze unbiasedness, and empirically evaluate.
result Proposed estimators can predict absolute values, rank, and optimize evaluated policies.
Study online learning with off-policy feedback in adversarial bandit problems.
problem Learning with limited direct feedback in sequential decision making.
method Proposed algorithms that adapt pessimistic reward estimators to handle unknown behavior policy.
result Guaranteed regret bounds scaling with policy mismatch, improving performance against well-covered comparators.
Novel LSE estimator improves off-policy learning and evaluation.
problem High variance and poor performance with low-quality propensity scores and heavy-tailed reward distributions.
method Introduces a novel estimator based on the log-sum-exponential (LSE) operator.
result Achieves convergence rate of O(n−ε/(1+ε)) for regret bounds. Boosting for off-policy learning reduces empirical risk.
problem Learning from logged bandit feedback without labeled data.
method A boosting algorithm optimizing policy's expected reward.
result Excess empirical risk decreases with each round of boosting.
Industrial recommender systems deal with extremely large action spaces -- many millions of items to recommend. Moreover, they need to serve billions of users, who are unique at any point in time, making a complex user state space. Luckily, huge quantities of logged implicit feedback (e.g., user clicks, dwell time) are …
In semantic parsing for question-answering, it is often too expensive to collect gold parses or even gold answers as supervision signals. We propose to convert model outputs into a set of human-understandable statements which allow non-expert users to act as proofreaders, providing error markings as learning signals to…
What is the most statistically efficient way to do off-policy evaluation and optimization with batch data from bandit feedback? For log data generated by contextual bandit algorithms, we consider offline estimators for the expected reward from a counterfactual policy. Our estimators are shown to have lowest variance in…
When learning from a batch of logged bandit feedback, the discrepancy between the policy to be learned and the off-policy training data imposes statistical and computational challenges. Unlike classical supervised learning and online learning settings, in batch contextual bandit learning, one only has access to a colle…
Paper tackles RLHF with DCPPO method, proving near-optimal suboptimality.
problem Challenges in offline RLHF with limited human feedback and bounded rationality.
method DCPPO method involving three stages: MLE, reward function recovery, and pessimistic value iteration.
result DCPPO's suboptimality almost matches classical pessimistic offline RL in terms of distribution shift and dimension.
Optimizes LLM prompts using logged user feedback.
problem Naive approaches to optimizing LLM prompts suffer from high variance and bias.
method Kernel-based off-policy gradient method leveraging sentence similarity.
result Substantially reduces variance and suppresses bias in optimizing prompts.
Proposes a new resampling method for off-policy evaluation in stochastic control.
problem Estimating policy performance from data generated under a different policy.
method K-nearest neighbor resampling procedure for off-policy evaluation.
result Statistical consistency results for the proposed method under weak conditions.
The ability to perform offline A/B-testing and off-policy learning using logged contextual bandit feedback is highly desirable in a broad range of applications, including recommender systems, search engines, ad placement, and personalized health care. Both offline A/B-testing and off-policy learning require a counterfa…
In machine learning we often try to optimise a decision rule that would have worked well over a historical dataset; this is the so called empirical risk minimisation principle. In the context of learning from recommender system logs, applying this principle becomes a problem because we do not have available the reward …
Paper offers a fast convergence theory for offline decision making.
problem Offline decision making problems, including reinforcement learning and off-policy evaluation.
method Introduces a framework (DMOF) and algorithm (EDD) with a fast convergence guarantee.
result Demonstrates a fast convergence guarantee with a lower bound complement.
Adapts GRPO for off-policy RL, improving reward.
problem Improving training stability and efficiency in RL.
method Adapts GRPO to off-policy setting, uses clipped surrogate objectives.
result Off-policy GRPO outperforms on-policy GRPO in empirical tests.
A new method for identifying the best arm in multi-armed bandits with mediator feedback.
problem Sequential decision-making with unknown mediator policies.
method Best-arm identification under mediators' feedback (BAI-MF) with statistical lower bound and sequential decision-making strategy.
result The proposed algorithm matches the lower bound on sample complexity for identifying the best arm.
This work studies the problem of batch off-policy evaluation for Reinforcement Learning in partially observable environments. Off-policy evaluation under partial observability is inherently prone to bias, with risk of arbitrarily large errors. We define the problem of off-policy evaluation for Partially Observable Mark…
New method improves off-policy critic evaluation in reinforcement learning.
problem High variance and instability in off-policy policy evaluation.
method Doubly robust estimators applied to actor-critic algorithms.
result Doubly robust estimation significantly improves performance in continuous control tasks.
Evolutionary Strategies optimize hyper-parameters for off-policy learning.
problem Hyper-parameter sensitivity in off-policy learning.
method Application of Evolutionary Strategies for online hyper-parameter tuning.
result Our method outperforms state-of-the-art baselines.
New method reduces state distribution mismatch in off-policy RL.
problem State distribution mismatch in off-policy RL algorithms.
method Develops a novel constrained off-policy gradient objective to minimize state distribution shift.
result Minimizing state distribution shift improves performance in off-policy RL algorithms.
This paper extends off-policy reinforcement learning to the multi-agent case in which a set of networked agents communicating with their neighbors according to a time-varying graph collaboratively evaluates and improves a target policy while following a distinct behavior policy. To this end, the paper develops a multi-…
Memory-efficient algorithm reduces variance in off-policy RL.
problem High variance in off-policy policy optimization.
method Memory-efficient, stochastically variance-reduced algorithm using off-policy samples.
result Empirically validated effectiveness of the proposed algorithm.
Paper improves bootstrapping for off-policy reinforcement learning inference.
problem Improving bootstrapping for off-policy reinforcement learning inference.
method Proposes a bootstrapping FQE method for off-policy statistical inference and a subsampling procedure to improve runtime.
result Asymptotically efficient and distributionally consistent bootstrapping FQE method for off-policy inference.
We propose and analyze an alternate approach to off-policy multi-step temporal difference learning, in which off-policy returns are corrected with the current Q-function in terms of rewards, rather than with the target policy in terms of transition probabilities. We prove that such approximate corrections are sufficien…
A new estimator improves off-policy evaluation in RL, outperforming existing methods.
problem Estimating performance of a new policy using historical data from a different policy.
method Doubly-robust estimator based on Targeted Maximum Likelihood Estimation, with variance reduction techniques.
result Our estimator uniformly outperforms existing methods across various RL environments and levels of model misspecification.
The paper introduces a novel method for stable off-policy learning using value function chaining.
problem Stability issues in off-policy reinforcement learning.
method The approach involves learning on-policy first, then chaining off-policy value estimates.
result The method guarantees convergence and can approximate off-policy TD solutions.
Temporal difference learning and Residual Gradient methods are the most widely used temporal difference based learning algorithms; however, it has been shown that none of their objective functions is optimal w.r.t approximating the true value function V. Two novel algorithms are proposed to approximate the true value…
RO-TD learns sparse value functions efficiently.
problem Learning sparse value functions efficiently.
method RO-TD integrates off-policy convergent gradient TD methods and online convex regularization.
result RO-TD learns sparse value functions with low computational complexity.
Stabilizes policy optimization with off-policy data using divergence augmentation.
problem Premature convergence and instability in policy optimization with off-policy data.
method Incorporates Bregman divergence between behavior and current policies to ensure safe policy updates.
result Empirically shows better performance in data-scarce scenarios compared to other algorithms.
A new approach for deep exploration in sparse reward reinforcement learning.
problem Slow or no learning in reinforcement learning with rare rewards.
method Long-term visitation count planning and decoupling exploration and exploitation.
result Significantly outperforms existing methods in sparse reward environments.
Monotonic policy improvement and off-policy learning are two main desirable properties for reinforcement learning algorithms. In this paper, by lower bounding the performance difference of two policies, we show that the monotonic policy improvement is guaranteed from on- and off-policy mixture samples. An optimization …
Proposes a convergent TD algorithm for off-policy RL.
problem Learning value function from different policies in RL.
method Convergent on-policy TD algorithm with linear function approximation.
result Proposes a convergent TD algorithm for off-policy RL.
The principal contribution of this paper is a conceptual framework for off-policy reinforcement learning, based on conditional expectations of importance sampling ratios. This framework yields new perspectives and understanding of existing off-policy algorithms, and reveals a broad space of unexplored algorithms. We th…
New algorithm for sequential off-policy learning improves performance over batch methods.
problem Training policies from logged interaction data in a sequential setting.
method Combines Logarithmic Smoothing with online PAC-Bayesian tools.
result Improves performance and accelerates convergence in sequential off-policy learning.
Paper proposes a framework for reliable off-policy evaluation in reinforcement learning.
problem Quantifying uncertainty in off-policy estimates for safe deployment of target policies.
method Distributionally robust optimization for creating confidence bounds.
result Non-asymptotic and asymptotic guarantees for robust cumulative reward estimates.
Off-policy reinforcement learning aims to leverage experience collected from prior policies for sample-efficient learning. However, in practice, commonly used off-policy approximate dynamic programming methods based on Q-learning and actor-critic methods are highly sensitive to the data distribution, and can make only …
New algorithms improve policy evaluation in reinforcement learning.
problem Off-policy stability and on-policy efficiency issues in policy evaluation.
method Introduced novel algorithms using oblique projection method.
result Demonstrated both off-policy stability and on-policy efficiency.
We study the problem of off-policy policy optimization in Markov decision processes, and develop a novel off-policy policy gradient method. Prior off-policy policy gradient approaches have generally ignored the mismatch between the distribution of states visited under the behavior policy used to collect data, and what …
Method selects best estimator for off-policy evaluation.
problem Choosing the best estimator for off-policy evaluation.
method Generic data-driven method for estimator selection.
result Method is competitive with oracle estimator, up to a constant factor.
OSIRIS reduces variance in off-policy evaluation by omitting irrelevant states.
problem High variance in importance sampling-based OPE estimators.
method OSIRIS reduces variance by omitting likelihood ratios associated with states irrelevant to return.
result OSIRIS is unbiased and has lower variance than ordinary importance sampling.
Develops confidence bounds for off-policy evaluation in contextual bandits.
problem Evaluating policies that were not used to collect data.
method Martingale analysis for non-asymptotic, non-parametric, and valid confidence sequences.
result Empirically tight bounds on failure probability and width.
Improves RL algorithms with two techniques.
problem Enhance off-policy RL performance.
method Formulates RL as proximal point iteration; uses value functions for improved action value estimate.
result Significant performance improvement on RL benchmarks.
New estimator uses clustering to improve off-policy evaluation accuracy.
problem Improving off-policy evaluation accuracy when logging and evaluation policies differ.
method Proposes an estimator that shares information across similar contexts using clustering.
result Clustering contexts improves estimation accuracy, especially in deficient information settings.
SUNRISE improves off-policy RL algorithms by integrating ensemble methods.
problem Stability and exploration issues in off-policy RL algorithms.
method SUNRISE combines ensemble-based weighted Bellman backups and upper-confidence bounds for efficient exploration.
result SUNRISE improves the performance of off-policy RL algorithms across various domains.
New algorithm reduces bias in off-policy reinforcement learning.
problem Challenges in designing off-policy reinforcement learning algorithms.
method Doubly robust off-policy actor-critic (DR-Off-PAC) with a single timescale structure.
result Establishes the first overall sample complexity analysis for a single time-scale off-policy AC algorithm.
HO2 learns options from data efficiently, improving robot manipulation tasks.
problem Learning options from raw pixel inputs in 3D robot manipulation tasks.
method HO2 infers likely option choices and trains all policy components off-policy.
result HO2 outperforms existing methods on 3D robot manipulation tasks.
This paper investigates the problem of online prediction learning, where learning proceeds continuously as the agent interacts with an environment. The predictions made by the agent are contingent on a particular way of behaving, represented as a value function. However, the behavior used to select actions and generate…
Paper addresses off-policy evaluation and learning with covariate shift.
problem Evaluating and training a new policy using historical data with a covariate shift.
method Derives efficiency bounds and proposes doubly robust estimators for OPE and OPL under covariate shift.
result Proposes estimators for off-policy evaluation and learning under covariate shift.