This study designs an adaptive experiment for efficiently estimating average treatment effect (ATEs). We consider an adaptive experiment where an experimenter sequentially samples an experimental unit from a covariate density decided by the experimenter and assigns a treatment. After assigning a treatment, the experimenter observes the corresponding outcome immediately. At the end of the experiment, the experimenter estimates an ATE using gathered samples. The objective of the experimenter is to estimate the ATE with a smaller asymptotic variance. Existing studies have designed experiments that adaptively optimize the propensity score (treatment-assignment probability). As a generalization of such an approach, we propose a framework under which an experimenter optimizes the covariate density, as well as the propensity score, and find that optimizing both covariate density and propensity score reduces the asymptotic variance more than optimizing only the propensity score. Based on this idea, in each round of our experiment, the experimenter optimizes the covariate density and propensity score based on past observations. To design an adaptive experiment, we first derive the efficient covariate density and propensity score that minimizes the semiparametric efficiency bound, a lower bound for the asymptotic variance given a fixed covariate density and a fixed propensity score. Next, we design an adaptive experiment using the efficient covariate density and propensity score sequentially estimated during the experiment. Lastly, we propose an ATE estimator whose asymptotic variance aligns with the minimized semiparametric efficiency bound.
This study considers the linear contextual bandit problem with independent and identically distributed (i.i.d.) contexts. In this problem, existing studies have proposed Best-of-Both-Worlds (BoBW) algorithms whose regrets satisfy $O(\log^2(T))$ for the number of rounds $T$ in a stochastic regime with a suboptimality gap lower-bounded by a positive constant, while satisfying $O(\sqrt{T})$ in an adversarial regime. However, the dependency on $T$ has room for improvement, and the suboptimality-gap assumption can be relaxed. For this issue, this study proposes an algorithm whose regret satisfies $O(\log(T))$ in the setting when the suboptimality gap is lower-bounded. Furthermore, we introduce a margin condition, a milder assumption on the suboptimality gap. That condition characterizes the problem difficulty linked to the suboptimality gap using a parameter $\beta \in (0, \infty]$. We then show that the algorithm's regret satisfies $O\left(\left\{\log(T)\right\}^{\frac{1+\beta}{2+\beta}}T^{\frac{1}{2+\beta}}\right)$. Here, $\beta= \infty$ corresponds to the case in the existing studies where a lower bound exists in the suboptimality gap, and our regret satisfies $O(\log(T))$ in that case. Our proposed algorithm is based on the Follow-The-Regularized-Leader with the Tsallis entropy and referred to as the $\alpha$-Linear-Contextual (LC)-Tsallis-INF.
This study investigates the estimation and the statistical inference about Conditional Average Treatment Effects (CATEs), which have garnered attention as a metric representing individualized causal effects. In our data-generating process, we assume linear models for the outcomes associated with binary treatments and define the CATE as a difference between the expected outcomes of these linear models. This study allows the linear models to be high-dimensional, and our interest lies in consistent estimation and statistical inference for the CATE. In high-dimensional linear regression, one typical approach is to assume sparsity. However, in our study, we do not assume sparsity directly. Instead, we consider sparsity only in the difference of the linear models. We first use a doubly robust estimator to approximate this difference and then regress the difference on covariates with Lasso regularization. Although this regression estimator is consistent for the CATE, we further reduce the bias using the techniques in double/debiased machine learning (DML) and debiased Lasso, leading to $\sqrt{n}$-consistency and confidence intervals. We refer to the debiased estimator as the triple/debiased Lasso (TDL), applying both DML and debiased Lasso techniques. We confirm the soundness of our proposed method through simulation studies.
We study a primal-dual reinforcement learning (RL) algorithm for the online constrained Markov decision processes (CMDP) problem, wherein the agent explores an optimal policy that maximizes return while satisfying constraints. Despite its widespread practical use, the existing theoretical literature on primal-dual RL algorithms for this problem only provides sublinear regret guarantees and fails to ensure convergence to optimal policies. In this paper, we introduce a novel policy gradient primal-dual algorithm with uniform probably approximate correctness (Uniform-PAC) guarantees, simultaneously ensuring convergence to optimal policies, sublinear regret, and polynomial sample complexity for any target accuracy. Notably, this represents the first Uniform-PAC algorithm for the online CMDP problem. In addition to the theoretical guarantees, we empirically demonstrate in a simple CMDP that our algorithm converges to optimal policies, while an existing algorithm exhibits oscillatory performance and constraint violation.
Evidence-based targeting has been a topic of growing interest among the practitioners of policy and business. Formulating decision-maker's policy learning as a fixed-budget best arm identification (BAI) problem with contextual information, we study an optimal adaptive experimental design for policy learning with multiple treatment arms. In the sampling stage, the planner assigns treatment arms adaptively over sequentially arriving experimental units upon observing their contextual information (covariates). After the experiment, the planner recommends an individualized assignment rule to the population. Setting the worst-case expected regret as the performance criterion of adaptive sampling and recommended policies, we derive its asymptotic lower bounds, and propose a strategy, Adaptive Sampling-Policy Learning strategy (PLAS), whose leading factor of the regret upper bound aligns with the lower bound as the size of experimental units increases.
This study investigates the problem of $K$-armed linear contextual bandits, an instance of the multi-armed bandit problem, under an adversarial corruption. At each round, a decision-maker observes an independent and identically distributed context and then selects an arm based on the context and past observations. After selecting an arm, the decision-maker incurs a loss corresponding to the selected arm. The decision-maker aims to minimize the cumulative loss over the trial. The goal of this study is to develop a strategy that is effective in both stochastic and adversarial environments, with theoretical guarantees. We first formulate the problem by introducing a novel setting of bandits with adversarial corruption, referred to as the contextual adversarial regime with a self-bounding constraint. We assume linear models for the relationship between the loss and the context. Then, we propose a strategy that extends the RealLinExp3 by Neu & Olkhovskaya (2020) and the Follow-The-Regularized-Leader (FTRL). The regret of our proposed algorithm is shown to be upper-bounded by $O\left(\min\left\{\frac{(\log(T))^3}{\Delta_{*}} + \sqrt{\frac{C(\log(T))^3}{\Delta_{*}}},\ \ \sqrt{T}(\log(T))^2\right\}\right)$, where $T \in\mathbb{N}$ is the number of rounds, $\Delta_{*} > 0$ is the constant minimum gap between the best and suboptimal arms for any context, and $C\in[0, T] $ is an adversarial corruption parameter. This regret upper bound implies $O\left(\frac{(\log(T))^3}{\Delta_{*}}\right)$ in a stochastic environment and by $O\left( \sqrt{T}(\log(T))^2\right)$ in an adversarial environment. We refer to our strategy as the Best-of-Both-Worlds (BoBW) RealFTRL, due to its theoretical guarantees in both stochastic and adversarial regimes.
We address the problem of best arm identification (BAI) with a fixed budget for two-armed Gaussian bandits. In BAI, given multiple arms, we aim to find the best arm, an arm with the highest expected reward, through an adaptive experiment. Kaufmann et al. (2016) develops a lower bound for the probability of misidentifying the best arm. They also propose a strategy, assuming that the variances of rewards are known, and show that it is asymptotically optimal in the sense that its probability of misidentification matches the lower bound as the budget approaches infinity. However, an asymptotically optimal strategy is unknown when the variances are unknown. For this open issue, we propose a strategy that estimates variances during an adaptive experiment and draws arms with a ratio of the estimated standard deviations. We refer to this strategy as the Neyman Allocation (NA)-Augmented Inverse Probability weighting (AIPW) strategy. We then demonstrate that this strategy is asymptotically optimal by showing that its probability of misidentification matches the lower bound when the budget approaches infinity, and the gap between the expected rewards of two arms approaches zero (small-gap regime). Our results suggest that under the worst-case scenario characterized by the small-gap regime, our strategy, which employs estimated variance, is asymptotically optimal even when the variances are unknown.
This study investigates the problem of identifying the best treatment arm, a treatment arm with the highest expected outcome. We aim to identify the best treatment arm with a lower probability of misidentification, which has been explored under various names across numerous research fields, including \emph{best arm identification} (BAI) and ordinal optimization. In our experiments, the number of treatment-allocation rounds is fixed. In each round, a decision-maker allocates a treatment arm to an experimental unit and observes a corresponding outcome, which follows a Gaussian distribution with a variance different among treatment arms. At the end of the experiment, we recommend one of the treatment arms as an estimate of the best treatment arm based on the observations. The objective of the decision-maker is to design an experiment that minimizes the probability of misidentifying the best treatment arm. With this objective in mind, we develop lower bounds for the probability of misidentification under the small-gap regime, where the gaps of the expected outcomes between the best and suboptimal treatment arms approach zero. Then, assuming that the variances are known, we design the Generalized-Neyman-Allocation (GNA)-empirical-best-arm (EBA) strategy, which is an extension of the Neyman allocation proposed by Neyman (1934) and the Uniform-EBA strategy proposed by Bubeck et al. (2011). For the GNA-EBA strategy, we show that the strategy is asymptotically optimal because its probability of misidentification aligns with the lower bounds as the sample size approaches infinity under the small-gap regime. We refer to such optimal strategies as locally asymptotic optimal because their performance aligns with the lower bounds within restricted situations characterized by the small-gap regime.
Consider a scenario where we have access to train data with both covariates and outcomes while test data only contains covariates. In this scenario, our primary aim is to predict the missing outcomes of the test data. With this objective in mind, we train parametric regression models under a covariate shift, where covariate distributions are different between the train and test data. For this problem, existing studies have proposed covariate shift adaptation via importance weighting using the density ratio. This approach averages the train data losses, each weighted by an estimated ratio of the covariate densities between the train and test data, to approximate the test-data risk. Although it allows us to obtain a test-data risk minimizer, its performance heavily relies on the accuracy of the density ratio estimation. Moreover, even if the density ratio can be consistently estimated, the estimation errors of the density ratio also yield bias in the estimators of the regression model's parameters of interest. To mitigate these challenges, we introduce a doubly robust estimator for covariate shift adaptation via importance weighting, which incorporates an additional estimator for the regression function. Leveraging double machine learning techniques, our estimator reduces the bias arising from the density ratio estimation errors. We demonstrate the asymptotic distribution of the regression parameter estimator. Notably, our estimator remains consistent if either the density ratio estimator or the regression function is consistent, showcasing its robustness against potential errors in density ratio estimation. Finally, we confirm the soundness of our proposed method via simulation studies.