Predicting Long Term Sequential Policy Value Using Softer Surrogates

Performing policy evaluation in education, healthcare and online commerce can be challenging, because it can require waiting substantial amounts of time to observe outcomes over the desired horizon of interest. While offline evaluation methods can be used to estimate the performance of a new decision policy from historical data in some cases, such methods struggle when the new policy involves novel actions or is being run in a new decision process with potentially different dynamics. Here we consider how to estimate the full-horizon value of a new decision policy using only short-horizon data from the new policy, and historical full-horizon data from a different behavior policy. We introduce two new estimators for this setting, including a doubly robust estimator, and provide formal analysis of their properties. Our empirical results on two realistic simulators, of HIV treatment and sepsis treatment, show that our methods can often provide informative estimates of a new decision policy ten times faster than waiting for the full horizon, highlighting that it may be possible to quickly identify if a new decision policy, involving new actions, is better or worse than existing past policies.

Automatic Doubly Robust Forests

This paper proposes the automatic Doubly Robust Random Forest (DRRF) algorithm for estimating the conditional expectation of a moment functional in the presence of high-dimensional nuisance functions. DRRF combines the automatic debiasing framework using the Riesz representer (Chernozhukov et al., 2022) with non-parametric, forest-based estimation methods for the conditional moment (Athey et al., 2019; Oprescu et al., 2019). In contrast to existing methods, DRRF does not require prior knowledge of the form of the debiasing term nor impose restrictive parametric or semi-parametric assumptions on the target quantity. Additionally, it is computationally efficient for making predictions at multiple query points and significantly reduces runtime compared to methods such as Orthogonal Random Forest (Oprescu et al., 2019). We establish the consistency and asymptotic normality results of DRRF estimator under general assumptions, allowing for the construction of valid confidence intervals. Through extensive simulations in heterogeneous treatment effect (HTE) estimation, we demonstrate the superior performance of DRRF over benchmark approaches in terms of estimation accuracy, robustness, and computational efficiency.

Personalized Adaptation via In-Context Preference Learning

Reinforcement Learning from Human Feedback (RLHF) is widely used to align Language Models (LMs) with human preferences. However, existing approaches often neglect individual user preferences, leading to suboptimal personalization. We present the Preference Pretrained Transformer (PPT), a novel approach for adaptive personalization using online user feedback. PPT leverages the in-context learning capabilities of transformers to dynamically adapt to individual preferences. Our approach consists of two phases: (1) an offline phase where we train a single policy model using a history-dependent loss function, and (2) an online phase where the model adapts to user preferences through in-context learning. We demonstrate PPT's effectiveness in a contextual bandit setting, showing that it achieves personalized adaptation superior to existing methods while significantly reducing the computational costs. Our results suggest the potential of in-context learning for scalable and efficient personalization in large language models.

Orthogonal Causal Calibration

Estimates of causal parameters such as conditional average treatment effects and conditional quantile treatment effects play an important role in real-world decision making. Given this importance, one should ensure these estimators are calibrated. While there is a rich literature on calibrating estimators of non-causal parameters, very few methods have been derived for calibrating estimators of causal parameters, or more generally estimators of quantities involving nuisance parameters. In this work, we provide a general framework for calibrating predictors involving nuisance estimation. We consider a notion of calibration defined with respect to an arbitrary, nuisance-dependent loss $\ell$, under which we say an estimator $\theta$ is calibrated if its predictions cannot be changed on any level set to decrease loss. We prove generic upper bounds on the calibration error of any causal parameter estimate $\theta$ with respect to any loss $\ell$ using a concept called Neyman Orthogonality. Our bounds involve two decoupled terms - one measuring the error in estimating the unknown nuisance parameters, and the other representing the calibration error in a hypothetical world where the learned nuisance estimates were true. We use our bound to analyze the convergence of two sample splitting algorithms for causal calibration. One algorithm, which applies to universally orthogonalizable loss functions, transforms the data into generalized pseudo-outcomes and applies an off-the-shelf calibration procedure. The other algorithm, which applies to conditionally orthogonalizable loss functions, extends the classical uniform mass binning algorithm to include nuisance estimation. Our results are exceedingly general, showing that essentially any existing calibration algorithm can be used in causal settings, with additional loss only arising from errors in nuisance estimation.

Consistency of Neural Causal Partial Identification

Recent progress in Neural Causal Models (NCMs) showcased how identification and partial identification of causal effects can be automatically carried out via training of neural generative models that respect the constraints encoded in a given causal graph [Xia et al. 2022, Balazadeh et al. 2022]. However, formal consistency of these methods has only been proven for the case of discrete variables or only for linear causal models. In this work, we prove consistency of partial identification via NCMs in a general setting with both continuous and categorical variables. Further, our results highlight the impact of the design of the underlying neural network architecture in terms of depth and connectivity as well as the importance of applying Lipschitz regularization in the training phase. In particular, we provide a counterexample showing that without Lipschitz regularization the NCM may not be asymptotically consistent. Our results are enabled by new results on the approximability of structural causal models via neural generative models, together with an analysis of the sample complexity of the resulting architectures and how that translates into an error in the constrained optimization problem that defines the partial identification bounds.

Direct Preference Optimization With Unobserved Preference Heterogeneity

RLHF has emerged as a pivotal step in aligning language models with human objectives and values. It typically involves learning a reward model from human preference data and then using reinforcement learning to update the generative model accordingly. Conversely, Direct Preference Optimization (DPO) directly optimizes the generative model with preference data, skipping reinforcement learning. However, both RLHF and DPO assume uniform preferences, overlooking the reality of diverse human annotators. This paper presents a new method to align generative models with varied human preferences. We propose an Expectation-Maximization adaptation to DPO, generating a mixture of models based on latent preference types of the annotators. We then introduce a min-max regret ensemble learning model to produce a single generative method to minimize worst-case regret among annotator subgroups with similar latent factors. Our algorithms leverage the simplicity of DPO while accommodating diverse preferences. Experimental results validate the effectiveness of our approach in producing equitable generative policies.

Uniform Inference for Subsampled Moment Regression

We propose a method for constructing a confidence region for the solution to a conditional moment equation. The method is built around a class of algorithms for nonparametric regression based on subsampled kernels. This class includes random forest regression. We bound the error in the confidence region's nominal coverage probability, under the restriction that the conditional moment equation of interest satisfies a local orthogonality condition. The method is applicable to the construction of confidence regions for conditional average treatment effects in randomized experiments, among many other similar problems encountered in applied economics and causal inference. As a by-product, we obtain several new order-explicit results on the concentration and normal approximation of high-dimensional $U$-statistics.

Taking a Moment for Distributional Robustness

A rich line of recent work has studied distributionally robust learning approaches that seek to learn a hypothesis that performs well, in the worst-case, on many different distributions over a population. We argue that although the most common approaches seek to minimize the worst-case loss over distributions, a more reasonable goal is to minimize the worst-case distance to the true conditional expectation of labels given each covariate. Focusing on the minmax loss objective can dramatically fail to output a solution minimizing the distance to the true conditional expectation when certain distributions contain high levels of label noise. We introduce a new min-max objective based on what is known as the adversarial moment violation and show that minimizing this objective is equivalent to minimizing the worst-case $\ell_2$-distance to the true conditional expectation if we take the adversary's strategy space to be sufficiently rich. Previous work has suggested minimizing the maximum regret over the worst-case distribution as a way to circumvent issues arising from differential noise levels. We show that in the case of square loss, minimizing the worst-case regret is also equivalent to minimizing the worst-case $\ell_2$-distance to the true conditional expectation. Although their objective and our objective both minimize the worst-case distance to the true conditional expectation, we show that our approach provides large empirical savings in computational cost in terms of the number of groups, while providing the same noise-oblivious worst-distribution guarantee as the minimax regret approach, thus making positive progress on an open question posed by Agarwal and Zhang (2022).

Dynamic Local Average Treatment Effects

We consider Dynamic Treatment Regimes (DTRs) with one sided non-compliance that arise in applications such as digital recommendations and adaptive medical trials. These are settings where decision makers encourage individuals to take treatments over time, but adapt encouragements based on previous encouragements, treatments, states, and outcomes. Importantly, individuals may choose to (not) comply with a treatment recommendation, whenever it is made available to them, based on unobserved confounding factors. We provide non-parametric identification, estimation, and inference for Dynamic Local Average Treatment Effects, which are expected values of multi-period treatment contrasts among appropriately defined complier subpopulations. Under standard assumptions in the Instrumental Variable and DTR literature, we show that one can identify local average effects of contrasts that correspond to offering treatment at any single time step. Under an additional cross-period effect-compliance independence assumption, which is satisfied in Staggered Adoption settings and a generalization of them, which we define as Staggered Compliance settings, we identify local average treatment effects of treating in multiple time periods.

Sequential Decision Making with Expert Demonstrations under Unobserved Heterogeneity

We study the problem of online sequential decision-making given auxiliary demonstrations from experts who made their decisions based on unobserved contextual information. These demonstrations can be viewed as solving related but slightly different tasks than what the learner faces. This setting arises in many application domains, such as self-driving cars, healthcare, and finance, where expert demonstrations are made using contextual information, which is not recorded in the data available to the learning agent. We model the problem as a zero-shot meta-reinforcement learning setting with an unknown task distribution and a Bayesian regret minimization objective, where the unobserved tasks are encoded as parameters with an unknown prior. We propose the Experts-as-Priors algorithm (ExPerior), a non-parametric empirical Bayes approach that utilizes the principle of maximum entropy to establish an informative prior over the learner's decision-making problem. This prior enables the application of any Bayesian approach for online decision-making, such as posterior sampling. We demonstrate that our strategy surpasses existing behaviour cloning and online algorithms for multi-armed bandits and reinforcement learning, showcasing the utility of our approach in leveraging expert demonstrations across different decision-making setups.