Off-policy evaluation (OPE) of ranking policies is challenging be- cause selecting and ordering multiple items from a candidate set makes the number of possible rankings grow combinatorially with the number of candidates and the ranking length. Consequently, Inverse Propensity Scoring (IPS), whose importance weight is the full-ranking probability ratio under the evaluation and logging policies, can have excessive variance. Independent IPS (IIPS) and Reward Interaction IPS (RIPS) reduce variance by imposing fixed assumptions on how users browse rankings, but may introduce bias when those assumptions mismatch actual behavior. Adaptive Inverse Propensity Scoring (AIPS) addresses this trade-off by adap- tively marginalizing importance weights over the actions that affect each position-wise reward. It attains minimum variance within a class of unbiased IPS-based estimators when the true user be- havior model is observed. However, its estimation accuracy may still degrade for longer rankings, and AIPS does not use a reward model for residual correction. We propose Adaptive Doubly Robust (ADR), which combines adaptive importance weighting with re- ward regression through a control-variate correction. We establish its unbiasedness when the true user behavior model is observed and characterize a sufficient condition under which it reduces vari- ance relative to AIPS. Across synthetic experiments with 10,000 simulations per condition, ADR improves mean squared error over AIPS and conventional ranking OPE estimators across a range of logged-data sizes and ranking lengths.
Generative recommenders increasingly emit semantic IDs (SIDs): each item is a short sequence of hierarchical discrete codes from a residual quantizer, decoded autoregressively. Before spending scarce A/B-test, a team may decide offline which decoder or reranking variants are worth testing - a job for off-policy evaluation (OPE). We ask a simple question: can the model's own SID tree serve as the action abstraction for that OPE? Our answer has three parts. (i) Under the near-argmax logging real recommenders use, per-item OPE is hopeless - as item-level effective sample size is usually small on production logs - but marginalizing items to code-prefix clusters restores estimable support and cuts error. (ii) This gain is thanks to coarsening, not to the hierarchy specifically; but the SID tree is what makes coarsening feasible in a generative system - each cluster's mass is exactly and cheaply returned by the decoder, whereas flat clustering requires enumerating item/leaf masses that a code-only decoder does not directly expose. (iii) Resolution depth is the operative knob - coarser under scarce support - and a conditional bias bound links the coarsening bias to the quantizer's worst-case reconstruction residual and the target-logging divergence.
We develop a new framework for flexible, nonlinear, and interpretable off-policy evaluation for infinite-horizon reinforcement learning. To handle large state spaces and support transparent decision-making, we model the Q-function using a nonlinear function class with a sparse additive structure. We derive high-probability finite-sample error bounds for estimating the value function of a target policy and show that the bounds depend only logarithmically on the ambient dimension $d$, thereby alleviating the curse of dimensionality. In contrast to most existing theory for off-policy evaluation, which typically assumes access to many trajectories, our analysis guarantees accurate value estimation when either the number of trajectories or the time horizon is sufficiently large. In addition, we propose a group-sparsity-based feature screening procedure that identifies, with high probability, a reduced feature set containing all relevant covariates. Numerical experiments demonstrate the effectiveness of the proposed approach.
Marc Pérez-Roig, David Fernández-Narro, Carlos Sáezcs.AI cs.LG
The dosing of intravenous fluids and vasopressors in sepsis is a sequential decision made under uncertainty and guided largely by clinical judgment, which makes it a natural target for reinforcement learning from historical care. Because a learned policy cannot be trialed on patients, its value must be estimated off-policy, and such estimates can be fragile and optimistic. This work advances the reliable evaluation of sepsis treatment policies by combining off-policy estimation, reliability diagnostics, and clinician-agreement analyses in a transparent validation framework. We modeled fluid and vasopressor dosing on a cohort of 36,872 septic ICU stays drawn from the MIMIC-IV critical-care database, as a discretized Markov decision process with 1,000 states and 25 actions, defined by a five-by-five grid of fluid and vasopressor levels and solved by policy iteration. The clinicians' behavior policy was estimated with a random forest, which mitigated the collapse of the Effective Sample Size (ESS 50.1 against 4.0 with smoothed counts) that otherwise destabilizes the importance-sampling estimate. The learned policy was evaluated with two estimators, weighted importance sampling (WIS) and fitted Q evaluation (FQE), with the ESS and clinician agreement as reliability checks. An empirical variable selection found that the composition of the state matters more than its size. Both estimators place the learned policy above the clinicians' return (WIS 50.8 and FQE 46.8 against 38.2, ESS 50.1), yet it departs only modestly from observed practice (total variation 0.18), favoring less intravenous fluid. These retrospective single-center off-policy results support the learned policy as a clinically plausible refinement of observed practice and motivate its further evaluation as a discordance-based clinical decision-support approach.
Mechanical ventilation is a critical life-support intervention, requiring dynamic adjustments to ventilator settings as a patient's condition evolves. While reinforcement learning (RL) offers a promising framework for optimizing these sequential decisions, standard approaches rely primarily on structured electronic health record (EHR) data, missing crucial clinical context recorded in free-text notes. Integrating longitudinal clinical notes into RL state spaces is challenging because notes are heavily inflated by temporal redundancy, such as copy-forward text, templating, and repetitive documentation, which dilutes time-local updates and degrades state representation quality. To address this, we propose a redundancy-aware multimodal state representation framework that explicitly removes duplicated note text over time before policy learning. We evaluate two computationally efficient temporal decomposition strategies for removing duplicated note text: (1) an embedding-space decomposition using singular value decomposition on local history subspaces, and (2) an interpretable sentence-level diff operation that filters out previously documented sentences before text encoding. Using real-world ICU data, we demonstrate that state representations constructed by stripping temporal note redundancy significantly outperform both structured-only and raw-note baselines across multiple off-policy evaluation methods (Model-Based Rollouts, Fitted Q-Evaluation, Weighted Importance Sampling, and Weighted Doubly Robust Evaluation). Our findings show that explicitly isolating new clinical information from repeated note text yields higher-quality state representations and directly improves RL performance for clinical decision support.
Organizations decide whom to treat under a budget and want to know what a targeting rule would have earned before deploying it. Off-policy evaluation promises this from logged data, but the deployable rule is a deterministic top-k policy: it removes all averaging over actions, so weak overlap hits the estimate directly. We benchmark six estimators across five datasets and two known-effect sweeps, and validate the mechanisms against a non-simulated paired reference. First, weak overlap is governed by logger-target action alignment, not by logging sharpness alone: what governs support is the logger's probability of the target's actions. Sharpening a logger built from the target's own score barely moves overlap over the tested range; action-level disagreement collapses it. Effective sample size ranks this risk across logging environments, but is weak at ranking candidates within the single log a practitioner holds, and its cut point does not transfer. Second, the optimizer's curse is not fixed by cross-fitting the outcome nuisance. When the rule is fit on the data used to evaluate it, cross-fitting the nuisance alone leaves the reuse bias in place and makes it worse. Honest policy-level splitting avoids the reuse by targeting the learning procedure's value -- a change of estimand, not a de-biasing of the full-sample policy. Third, propensity-estimation error is the largest degradation we measure: an out-of-fold estimate hurts IPS more than any other stress we apply, leaves doubly-robust estimation almost unchanged, and can invert the overlap diagnostic itself. Logging is synthesized and propensities floored at 0.02, so every failure occurs with bounded weights; the floor also reduces the two tuned hybrids to their untuned parents, leaving four practically distinct estimators, and all exact-value surfaces are synthetic or semi-synthetic. We release the benchmark; public data only.
Sang Su Lee, Vineeth Loganathan, Shishir Dash +1cs.LG
Personalizing marketing messages with contextual multi-armed bandits (CMABs) drives real business value, yet the objective that ultimately matters - a downstream conversion - is observed only weeks later, too late to drive online learning. Teams therefore train the bandit on a fast proxy reward, and separately must judge whether a contextual bandit is worth its complexity over sending one best message. Settling both decisions with the usual offline checks - a batch off-policy estimate, a marginal arm-discrimination test, a confidence interval - can mislead systematically under delayed feedback. We give an ordered diagnostic protocol that screens a reward-and-policy candidate on two axes, alignment (does optimizing the reward move the north-star?) and learnability (can the bandit identify the reward-optimal policy?), before trusting any reported lift. We validate it where the truth is known - a public off-policy-evaluation benchmark and a controllable synthetic generator - and illustrate it on a deployed large-marketplace push system (where, with five arms and one split, the evidence is directional rather than powered). Two lessons recur. (N1) A single offline number can mis-rank rewards: a denser reward signal gives the bandit more to learn from, so rewards that look tied in a static estimate pull apart once learning happens online. (N2) If you cannot tell in advance which single message is best, a per-user policy partly just avoids betting on the wrong one - that looks like personalization but is really robustness, so a "personalization premium" is easily overstated. Our contribution is methodological rather than algorithmic: the ordered protocol, the two lessons it surfaces, and the end-to-end experience of applying it to a delayed-feedback CMAB.
Standard offline reinforcement learning (RL) algorithms typically assume that the actions in the dataset are observed without error. However, in many real-world applications, the true actions are unobserved and only noisy proxies are available, causing existing RL methods to yield biased and potentially misleading conclusions. We study off-policy evaluation in infinite-horizon discounted Markov decision processes with hidden actions. By leveraging the next-state variable as a natural proxy for the unobserved action, we establish identification of the policy value and propose an influence-function-based estimator called LURE (Learning from the Unseen: Robust Estimator). LURE is multiply robust, remaining consistent under several combinations of correctly specified nuisance components, and is asymptotically normal, enabling valid statistical inference. To our knowledge, this is the first work to address offline RL with hidden actions. We demonstrate LURE's effectiveness through simulations and a sepsis management application using the MIMIC-III database.
Off-Policy Evaluation and Learning (OPE/L) in contextual bandits is rapidly gaining popularity in real systems because new policies can be evaluated and learned securely using only historical logged data. However, existing methods in OPE/L cannot handle many challenging but prevalent scenarios such as few-shot data, deterministic logging policies, and new actions. In many applications, such as personalized medicine, content recommendations, education, and advertising, we need to evaluate and learn new policies in the presence of these challenges. Existing methods cannot evaluate and optimize effectively in these situations due to the notorious variance issue or limited exploration in the logged data. To enable OPE/L even under these unsolved challenges, we propose a new problem setup of Cross-Domain OPE/L, where we have access not only to the logged data from the target domain in which the new policy will be implemented but also to logged datasets collected from other domains. This novel formulation is widely applicable because we can often use historical data not only from the target hospital, country, device, or user segment but also from other hospitals, countries, devices, or segments. We develop a new estimator and policy gradient method to solve OPE/L by leveraging both target and source datasets, resulting in substantially enhanced OPE/L in the previously unsolved situations in our empirical evaluations.
Joshua Spear, Matthieu Komorowski, Rebecca Pope +2cs.LG
This article presents a novel estimator for performing off-policy evaluation using only offline data for contextual bandits. The proposed estimator, Kernel-WIS is demonstrated to be asymptotically consistent and to empirically outperform strong baselines (including vanilla weighted importance sampling), particularly under complex conditions including behaviour policy miss-specification. The benefit of Kernel-WIS is derived from combining the bounded property of vanilla weighted importance sampling with the linearity of vanilla importance sampling.
Occupancy ratios correct distribution shift in offline reinforcement learning and are central to off-policy evaluation. Existing primal-dual and minimax methods typically estimate these ratios by enforcing occupancy-balance moments over a critic class. We propose fitted occupancy-ratio evaluation (FORE), a fitted fixed-point method that characterizes the discounted occupancy ratio through an adjoint Bellman recursion. At each iteration, FORE solves a single-level density-ratio objective on one-step-transition data, thereby projecting the adjoint Bellman image onto a log-ratio class in Kullback--Leibler (KL) divergence. Unlike analyses of fitted Q-evaluation, which typically require value-function realizability together with Bellman completeness or projected-operator stability, our central approximation condition is just realizability of the discounted occupancy ratio itself. Under this condition, the population KL-projected recursion contracts in relative entropy toward the true ratio by virtue of the adjoint Bellman operator being a KL-contraction. For the empirical recursion, we establish finite-sample regret bounds that yield convergence in KL up to log-ratio approximation error and a statistical error governed by the complexity of the ratio hypothesis class. The fitted ratio supports direct value estimation by reward reweighting, occupancy-weighted fitted Q-evaluation, and doubly robust estimation that combines the fitted ratio with a fitted Q-function. Together, these results identify discounted occupancy-ratio realizability as a sufficient condition for offline policy evaluation without any completeness assumptions.
Optimizing instructional policies in Intelligent Tutoring Systems (ITS) typically requires costly online experimentation or student simulators that may fail to capture real-world dynamics. This paper introduces an offline contextual bandit framework that learns new adaptive policies directly from logged interaction data. By mapping student-item interactions onto a continuous latent proficiency-difficulty scale using a Rasch model, we cast the tutoring process as a continuous stochastic bandit problem. We propose a novel reward function designed to optimize ''flow'' by balancing task challenge with student success. Our approach includes a round-specific behavior policy estimation that serves as both a propensity model for off-policy evaluation and a diagnostic tool for ITS adaptivity. We demonstrate the efficacy of this framework across four large-scale real-world datasets, achieving consistent policy improvements over the logged behavior policy. The results show that effective instructional policies can be learned and visualized within seconds of computation, providing a scalable path for improving adaptive learning systems without further data collection.
In offline Reinforcement Learning, immediate rewards in logged batch data are often unobserved due to sparse or irregular record-keeping, or censored beyond certain reward values. This issue arises in practical settings, including health care and marketing. We investigate off-policy evaluation (OPE) in finite-horizon Markov decision processes when rewards are missing not at random (MNAR), which breaks ignorability and induces selection bias even after conditioning on states and actions. To address this, we formalize a reward-dependent propensity model and use future states as shadow variables to identify the full-data conditional mean reward. We further introduce a bridge function that recovers the conditional mean reward without explicitly modeling the MNAR mechanism, and estimate it via a min-max procedure to avoid double sampling. Building upon these identification results, we propose an Fitted-Q-Evaluation-style estimator that propagates the recovered rewards while allowing target policies to depend on past missingness indicators. Finally, we establish consistency and finite-sample error bounds for our OPE estimator, and show through experiments the strong performance of our method compared to existing methods on simulated and MIMIC-III Sepsis data.
Kiet Q. H. Vo, Abbavaram Gowtham Reddy, Julian Rodemann +2cs.AI
We study off-policy evaluation (OPE) under strategic behavior where decision subjects (or agents) respond to a decision maker's policy by strategically modifying their covariates. Such behavior induces a policy-dependent covariate shift, breaking the standard assumption in existing methods that covariates are exogenous to the policy. Related work addresses this challenge by imposing strong assumptions such as repeated interactions or full knowledge of agents' response behavior, substantially limiting its applicability to OPE. In contrast, we consider a one-shot OPE setting where the decision maker has only partial knowledge of the agents' response behavior. Our key insight is that disclosing local information through post-hoc explanations reveals agents' pre-strategic covariates prior to adaptation, mitigating the information loss induced by strategic behavior. Leveraging this structure, we estimate a statistical model for the agents' responses and construct a doubly robust estimator for policy value. By assuming that the agents' cost sensitivity follows a conditional log-normal distribution, we establish consistency of the proposed estimator and validate our approach empirically. More broadly, our results highlight how interaction design can mitigate information asymmetry by revealing otherwise hidden structure in agents' strategic responses.
Evaluating large language model (LLM) agents in multi-turn interactive environments is expensive and risky, as it requires online environment interaction. We propose ADWM (Autoregressive Diffusion World Model), an evaluation framework that estimates the performance of a new LLM agent policy purely from pre-collected trajectories. The core idea is to learn a latent diffusion world model that simulates how the environment responds to the evaluation policy, without ever executing it in the real environment. Existing diffusion-based OPE methods guide full trajectories in a single pass by jointly diffusing states and actions, an assumption that breaks down for LLM agents whose actions are discrete text that must be sampled from the policy after observing the environment. Unlike autoregressive world models that suffer from compounding errors, ADWM models each transition as an independent denoising process, enabling reliable step-by-step rollouts where the world model and agent alternate in causal order. Crucially, the LLM agent under evaluation directly guides the diffusion generation at each step via a policy-conditioned score function, ensuring that simulated trajectories accurately reflect its decision-making patterns. Empirically, ADWM achieves accurate value estimates and evaluation reliability across diverse multi-turn agent tasks, demonstrating its promise as a practical framework for offline LLM agent evaluation.