Off-policy Evaluation Research Articles

Reinforcement Learning to Optimize Ventilator Settings for Patients on Invasive Mechanical Ventilation: Retrospective Study.

One of the significant changes in intensive care medicine over the past 2 decades is the acknowledgment that improper mechanical ventilation settings substantially contribute to pulmonary injury in critically ill patients. Artificial intelligence (AI) solutions can optimize mechanical ventilation settings in intensive care units (ICUs) and improve patient outcomes. Specifically, machine learning algorithms can be trained on large datasets of patient information and mechanical ventilation settings. These algorithms can then predict patient responses to different ventilation strategies and suggest personalized ventilation settings for individual patients. In this study, we aimed to design and evaluate an AI solution that could tailor an optimal ventilator strategy for each critically ill patient who requires mechanical ventilation. We proposed a reinforcement learning-based AI solution using observational data from multiple ICUs in the United States. The primary outcome was hospital mortality. Secondary outcomes were the proportion of optimal oxygen saturation and the proportion of optimal mean arterial blood pressure. We trained our AI agent to recommend low, medium, and high levels of 3 ventilator settings-positive end-expiratory pressure, fraction of inspired oxygen, and ideal body weight-adjusted tidal volume-according to patients' health conditions. We defined a policy as rules guiding ventilator setting changes given specific clinical scenarios. Off-policy evaluation metrics were applied to evaluate the AI policy. We studied 21,595 and 5105 patients' ICU stays from the e-Intensive Care Unit Collaborative Research (eICU) and Medical Information Mart for Intensive Care IV (MIMIC-IV) databases, respectively. Using the learned AI policy, we estimated the hospital mortality rate (eICU 12.1%, SD 3.1%; MIMIC-IV 29.1%, SD 0.9%), the proportion of optimal oxygen saturation (eICU 58.7%, SD 4.7%; MIMIC-IV 49%, SD 1%), and the proportion of optimal mean arterial blood pressure (eICU 31.1%, SD 4.5%; MIMIC-IV 41.2%, SD 1%). Based on multiple quantitative and qualitative evaluation metrics, our proposed AI solution outperformed observed clinical practice. Our study found that customizing ventilation settings for individual patients led to lower estimated hospital mortality rates compared to actual rates. This highlights the potential effectiveness of using reinforcement learning methodology to develop AI models that analyze complex clinical data for optimizing treatment parameters. Additionally, our findings suggest the integration of this model into a clinical decision support system for refining ventilation settings, supporting the need for prospective validation trials.

Off-Policy Evaluation in Doubly Inhomogeneous Environments

– This work aims to study off-policy evaluation (OPE) under scenarios where two key reinforcement learning (RL) assumptions—temporal stationarity and individual homogeneity are both violated. To handle the “double inhomogeneities”, we propose a class of latent factor models for the reward and transition functions, under which we develop a general OPE framework that consists of both model-based and model-free approaches. To our knowledge, this is the first article that develops statistically sound OPE methods in offline RL with double inhomogeneities. It contributes to a deeper understanding of OPE in environments, where standard RL assumptions are not met, and provides several practical approaches in these settings. We establish the theoretical properties of the proposed value estimators and empirically show that our approach outperforms state-of-the-art methods. Finally, we illustrate our method on a dataset from the Medical Information Mart for Intensive Care. An R implementation of the proposed procedure is available at https://github.com/ZeyuBian/2FEOPE . Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work.

Anytime-valid off-policy Inference for Contextual Bandits

Contextual bandit algorithms are ubiquitous tools for active sequential experimentation in healthcare and the tech industry. They involve online learning algorithms that adaptively learn policies over time to map observed contexts X t to actions A t in an attempt to maximize stochastic rewards R t . This adaptivity raises interesting but hard statistical inference questions, especially counterfactual ones: for example, it is often of interest to estimate the properties of a hypothetical policy that is different from the logging policy that was used to collect the data—a problem known as “off-policy evaluation” (OPE). Using modern martingale techniques, we present a comprehensive framework for OPE inference that relaxes unnecessary conditions made in some past works (such as performing inference at prespecified sample sizes, uniformly bounded importance weights, constant logging policies, and constant policy values, among others), significantly improving on them both theoretically and empirically. Importantly, our methods can be employed while the original experiment is still running (that is, not necessarily post hoc), when the logging policy may be itself changing (due to learning), and even if the context distributions are a highly dependent time series (such as if they are drifting over time). More concretely, we derive confidence sequences for various functionals of interest in OPE. These include doubly robust ones for time-varying off-policy mean reward values, but also confidence bands for the entire cumulative distribution function of the off-policy reward distribution. All of our methods (a) are valid at arbitrary stopping times; (b) only make nonparametric assumptions; (c) do not require importance weights to be uniformly bounded, and if they are, we do not need to know these bounds; and (d) adapt to the empirical variance of our estimators. In summary, our methods enable anytime-valid off-policy inference using adaptively collected contextual bandit data.

Reinforcement learning for intensive care medicine: actionable clinical insights from novel approaches to reward shaping and off-policy model evaluation

BackgroundReinforcement learning (RL) holds great promise for intensive care medicine given the abundant availability of data and frequent sequential decision-making. But despite the emergence of promising algorithms, RL driven bedside clinical decision support is still far from reality. Major challenges include trust and safety. To help address these issues, we introduce cross off-policy evaluation and policy restriction and show how detailed policy analysis may increase clinical interpretability. As an example, we apply these in the setting of RL to optimise ventilator settings in intubated covid-19 patients.MethodsWith data from the Dutch ICU Data Warehouse and using an exhaustive hyperparameter grid search, we identified an optimal set of Dueling Double-Deep Q Network RL models. The state space comprised ventilator, medication, and clinical data. The action space focused on positive end-expiratory pressure (peep) and fraction of inspired oxygen (FiO2) concentration. We used gas exchange indices as interim rewards, and mortality and state duration as final rewards. We designed a novel evaluation method called cross off-policy evaluation (OPE) to assess the efficacy of models under varying weightings between the interim and terminal reward components. In addition, we implemented policy restriction to prevent potentially hazardous model actions. We introduce delta-Q to compare physician versus policy action quality and in-depth policy inspection using visualisations.ResultsWe created trajectories for 1118 intensive care unit (ICU) admissions and trained 69,120 models using 8 model architectures with 128 hyperparameter combinations. For each model, policy restrictions were applied. In the first evaluation step, 17,182/138,240 policies had good performance, but cross-OPE revealed suboptimal performance for 44% of those by varying the reward function used for evaluation. Clinical policy inspection facilitated assessment of action decisions for individual patients, including identification of action space regions that may benefit most from optimisation.ConclusionCross-OPE can serve as a robust evaluation framework for safe RL model implementation by identifying policies with good generalisability. Policy restriction helps prevent potentially unsafe model recommendations. Finally, the novel delta-Q metric can be used to operationalise RL models in clinical practice. Our findings offer a promising pathway towards application of RL in intensive care medicine and beyond.

Get a Head Start: On-Demand Pedagogical Policy Selection in Intelligent Tutoring

Reinforcement learning (RL) is broadly employed in human-involved systems to enhance human outcomes. Off-policy evaluation (OPE) has been pivotal for RL in those realms since online policy learning and evaluation can be high-stake. Intelligent tutoring has raised tremendous attentions as highly challenging when applying OPE to human-involved systems, due to that students' subgroups can favor different pedagogical policies and the costly procedure that policies have to be induced fully offline and then directly deployed to the upcoming semester. In this work, we formulate on-demand pedagogical policy selection (ODPS) to tackle the challenges for OPE in intelligent tutoring. We propose a pipeline, EduPlanner, as a concrete solution for ODPS. Our pipeline results in an theoretically unbiased estimator, and enables efficient and customized policy selection by identifying subgroups over both historical data and on-arrival initial logs. We evaluate our approach on the Probability ITS that has been used in real classrooms for over eight years. Our study shows significant improvement on learning outcomes of students with EduPlanner, especially for the ones associated with low-performing subgroups.

Probabilistic Offline Policy Ranking with Approximate Bayesian Computation

In practice, it is essential to compare and rank candidate policies offline before real-world deployment for safety and reliability. Prior work seeks to solve this offline policy ranking (OPR) problem through value-based methods, such as Off-policy evaluation (OPE). However, they fail to analyze special case performance (e.g., worst or best cases), due to the lack of holistic characterization of policies’ performance. It is even more difficult to estimate precise policy values when the reward is not fully accessible under sparse settings. In this paper, we present Probabilistic Offline Policy Ranking (POPR), a framework to address OPR problems by leveraging expert data to characterize the probability of a candidate policy behaving like experts, and approximating its entire performance posterior distribution to help with ranking. POPR does not rely on value estimation, and the derived performance posterior can be used to distinguish candidates in worst-, best-, and average-cases. To estimate the posterior, we propose POPR-EABC, an Energy-based Approximate Bayesian Computation (ABC) method conducting likelihood-free inference. POPR-EABC reduces the heuristic nature of ABC by a smooth energy function, and improves the sampling efficiency by a pseudo-likelihood. We empirically demonstrate that POPR-EABC is adequate for evaluating policies in both discrete and continuous action spaces across various experiment environments, and facilitates probabilistic comparisons of candidate policies before deployment.

Distributional Off-Policy Evaluation for Slate Recommendations

Recommendation strategies are typically evaluated by using previously logged data, employing off-policy evaluation methods to estimate their expected performance. However, for strategies that present users with slates of multiple items, the resulting combinatorial action space renders many of these methods impractical. Prior work has developed estimators that leverage the structure in slates to estimate the expected off-policy performance, but the estimation of the entire performance distribution remains elusive. Estimating the complete distribution allows for a more comprehensive evaluation of recommendation strategies, particularly along the axes of risk and fairness that employ metrics computable from the distribution. In this paper, we propose an estimator for the complete off-policy performance distribution for slates and establish conditions under which the estimator is unbiased and consistent. This builds upon prior work on off-policy evaluation for slates and off-policy distribution estimation in reinforcement learning. We validate the efficacy of our method empirically on synthetic data as well as on a slate recommendation simulator constructed from real-world data (MovieLens-20M). Our results show a significant reduction in estimation variance and improved sample efficiency over prior work across a range of slate structures.

A multiagent reinforcement learning framework for off-policy evaluation in two-sided markets

A multiagent reinforcement learning framework for off-policy evaluation in two-sided markets

Off-policy evaluation for tabular reinforcement learning with synthetic trajectories

Off-policy evaluation for tabular reinforcement learning with synthetic trajectories

Offline Deep Reinforcement Learning and Off-Policy Evaluation for Personalized Basal Insulin Control in Type 1 Diabetes.

Recent advancements in hybrid closed-loop systems, also known as the artificial pancreas (AP), have been shown to optimize glucose control and reduce the self-management burdens for people living with type 1 diabetes (T1D). AP systems can adjust the basal infusion rates of insulin pumps, facilitated by real-time communication with continuous glucose monitoring. Deep reinforcement learning (DRL) has introduced new paradigms of basal insulin control algorithms. However, all the existing DRL-based AP controllers require extensive random online interactions between the agent and environment. While this can be validated in T1D simulators, it becomes impractical in real-world clinical settings. To this end, we propose an offline DRL framework that can develop and validate models for basal insulin control entirely offline. It comprises a DRL model based on the twin delayed deep deterministic policy gradient and behavior cloning, as well as off-policy evaluation (OPE) using fitted Q evaluation. We evaluated the proposed framework on an in silico dataset generated by the UVA/Padova T1D simulator, and the OhioT1DM dataset, a real clinical dataset. The performance on the in silico dataset shows that the offline DRL algorithm significantly increased time in range while reducing time below range and time above range for both adult and adolescent groups. Then, we used the OPE to estimate model performance on the clinical dataset, where a notable increase in policy values was observed for each subject. The results demonstrate that the proposed framework is a viable and safe method for improving personalized basal insulin control in T1D.

Efficient evaluation of natural stochastic policies in off-line reinforcement learning

Summary We study the efficient off-policy evaluation of natural stochastic policies, which are defined in terms of deviations from the unknown behaviour policy. This is a departure from the literature on off-policy evaluation that largely considers the evaluation of explicitly specified policies. Crucially, off-line reinforcement learning with natural stochastic policies can help alleviate issues of weak overlap, lead to policies that build upon current practice and improve policies’ implementability in practice. Compared with the classic case of a prespecified evaluation policy, when evaluating natural stochastic policies, the efficiency bound, which measures the best-achievable estimation error, is inflated since the evaluation policy itself is unknown. In this paper we derive the efficiency bounds of two major types of natural stochastic policies: tilting policies and modified treatment policies. We then propose efficient nonparametric estimators that attain the efficiency bounds under lax conditions and enjoy a partial double robustness property.

Proximal Reinforcement Learning: Efficient Off-Policy Evaluation in Partially Observed Markov Decision Processes

In applications of offline reinforcement learning to observational data, such as in healthcare or education, a general concern is that observed actions might be affected by unobserved factors, inducing confounding and biasing estimates derived assuming a perfect Markov decision process (MDP) model. In “Proximal Reinforcement Learning: Efficient Off-Policy Evaluation in Partially Observed Markov Decision Processes,” A. Bennett and N. Kallus tackle this by considering off-policy evaluation in a partially observed MDP (POMDP). Specifically, they consider estimating the value of a given target policy in an unknown POMDP, given observations of trajectories generated by a different and unknown policy, which may depend on the unobserved states. They consider both when the target policy value can be identified the observed data and, given identification, how best to estimate it. Both these problems are addressed by extending the framework of proximal causal inference to POMDP settings, using sequences of so-called bridge functions. This results in a novel framework for off-policy evaluation in POMDPs that they term proximal reinforcement learning, which they validate in various empirical settings.

Off-policy evaluation in partially observed Markov decision processes under sequential ignorability

Off-policy evaluation in partially observed Markov decision processes under sequential ignorability

Optimal discharge of patients from intensive care via a data-driven policy learning framework

Clinical decision support tools rooted in machine learning and optimization can provide significant value to healthcare providers through better management of intensive care units. In particular, it is important that intensive care unit patient discharge decisions account for the nuanced trade-off between decreasing the length of stay and the risk of readmission or death after discharge of a patient. This work introduces a comprehensive framework (i.e., not geared towards any particular disease or condition) for capturing this trade-off and to recommend optimal discharge timing decisions given the electronic health records of a patient. A data-driven approach is used to derive a parsimonious, discrete state space representation to represent the physiological condition of a given patient. Based on this model and a given cost function, an infinite-horizon discounted Markov decision process is formulated and solved numerically to compute an optimal discharge policy, whose performance is assessed using off-policy evaluation strategies. Extensive numerical experiments are performed to validate the proposed framework using real-life intensive care unit patient data.

Scaling Marginalized Importance Sampling to High-Dimensional State-Spaces via State Abstraction

We consider the problem of off-policy evaluation (OPE) in reinforcement learning (RL), where the goal is to estimate the performance of an evaluation policy, pie, using a fixed dataset, D, collected by one or more policies that may be different from pie. Current OPE algorithms may produce poor OPE estimates under policy distribution shift i.e., when the probability of a particular state-action pair occurring under pie is very different from the probability of that same pair occurring in D. In this work, we propose to improve the accuracy of OPE estimators by projecting the high-dimensional state-space into a low-dimensional state-space using concepts from the state abstraction literature. Specifically, we consider marginalized importance sampling (MIS) OPE algorithms which compute state-action distribution correction ratios to produce their OPE estimate. In the original ground state-space, these ratios may have high variance which may lead to high variance OPE. However, we prove that in the lower-dimensional abstract state-space the ratios can have lower variance resulting in lower variance OPE. We then highlight the challenges that arise when estimating the abstract ratios from data, identify sufficient conditions to overcome these issues, and present a minimax optimization problem whose solution yields these abstract ratios. Finally, our empirical evaluation on difficult, high-dimensional state-space OPE tasks shows that the abstract ratios can make MIS OPE estimators achieve lower mean-squared error and more robust to hyperparameter tuning than the ground ratios.

Policy-Adaptive Estimator Selection for Off-Policy Evaluation

Off-policy evaluation (OPE) aims to accurately evaluate the performance of counterfactual policies using only offline logged data. Although many estimators have been developed, there is no single estimator that dominates the others, because the estimators' accuracy can vary greatly depending on a given OPE task such as the evaluation policy, number of actions, and noise level. Thus, the data-driven estimator selection problem is becoming increasingly important and can have a significant impact on the accuracy of OPE. However, identifying the most accurate estimator using only the logged data is quite challenging because the ground-truth estimation accuracy of estimators is generally unavailable. This paper thus studies this challenging problem of estimator selection for OPE for the first time. In particular, we enable an estimator selection that is adaptive to a given OPE task, by appropriately subsampling available logged data and constructing pseudo policies useful for the underlying estimator selection task. Comprehensive experiments on both synthetic and real-world company data demonstrate that the proposed procedure substantially improves the estimator selection compared to a non-adaptive heuristic. Note that complete version with technical appendix is available on arXiv: http://arxiv.org/abs/2211.13904.

Catoni-style confidence sequences for heavy-tailed mean estimation

A confidence sequence (CS) is a sequence of confidence intervals that is valid at arbitrary data-dependent stopping times. These are useful in applications like A/B testing, multi-armed bandits, off-policy evaluation, election auditing, etc. We present three approaches to constructing a confidence sequence for the population mean, under the minimal assumption that only an upper bound σ2 on the variance is known. While previous works rely on light-tail assumptions like boundedness or subGaussianity (under which all moments of a distribution exist), the confidence sequences in our work are able to handle data from a wide range of heavy-tailed distributions. The best among our three methods — the Catoni-style confidence sequence — performs remarkably well in practice, essentially matching the state-of-the-art methods for σ2-subGaussian data, and provably attains the loglogt/t lower bound due to the law of the iterated logarithm. Our findings have important implications for sequential experimentation with unbounded observations, since the σ2-bounded-variance assumption is more realistic and easier to verify than σ2-subGaussianity (which implies the former). We also extend our methods to data with infinite variance, but having pth central moment (1<p<2).

Off-Policy Evaluation With Online Adaptation for Robot Exploration in Challenging Environments

Autonomous exploration has many important applications. However, classic information gain-based or frontier-based exploration only relies on the robot current state to determine the immediate exploration goal, which lacks the capability of predicting the value of future states and thus leads to inefficient exploration decisions. This letter presents a method to learn how “good” states are, measured by the state value function, to provide a guidance for robot exploration in real-world challenging environments. We formulate our work as an off-policy evaluation (OPE) problem for robot exploration (OPERE). It consists of offline Monte-Carlo training on real-world data and performs Temporal Difference (TD) online adaptation to optimize the trained value estimator. We also design an intrinsic reward function based on sensor information coverage to enable the robot to gain more information with sparse extrinsic rewards. Results show that our method enables the robot to predict the value of future states so as to better guide robot exploration. The proposed algorithm achieves better prediction and exploration performance compared with the state-of-the-arts. To the best of our knowledge, this work for the first time demonstrates value function prediction on real-world dataset for robot exploration in challenging subterranean and urban environments.

Temporal-difference emphasis learning with regularized correction for off-policy evaluation and control

Temporal-difference emphasis learning with regularized correction for off-policy evaluation and control

Towards more efficient and robust evaluation of sepsis treatment with deep reinforcement learning

BackgroundIn recent years, several studies have applied advanced AI methods, i.e., deep reinforcement learning, in discovering more efficient treatment policies for sepsis. However, due to a paucity of understanding of sepsis itself, the existing approaches still face a severe evaluation challenge, that is, how to properly evaluate the goodness of treatments during the learning process and the effectiveness of the final learned treatment policies.MethodsWe propose a deep inverse reinforcement learning with mini-tree model that integrates different aspects of factors into the reward formulation, including the critical factors in causing mortality and the key indicators in the existing sepsis treatment guidelines, in order to provide a more comprehensive evaluation of treatments during learning. A new off-policy evaluation method is then proposed to enable more robust evaluation of the learned policies by considering the weighted averaged value functions estimated until the current step.ResultsResults in the MIMIC-III dataset show that the proposed methods can achieve more efficient treatment policies with higher reliability compared to those used by the clinicians.ConclusionsA more sound and comprehensive evaluation of treatments of sepsis should consider the most critical factors in infulencing the mortality during treatment as well as those key indicators in the existing sepsis diagnosis guidelines.