Searcharxiv⌕ Search

arXiv subjects

Michele Santacatterina

Publications and source records attributed to Michele Santacatterina.

At least 19 recordsLinked to original sources

Plausible Videos Can Lead to Poor Decisions: Diagnosing and CURE-ing World Models for Planning

Action-conditioned world models are increasingly used as world models for planning, yet a world model that makes plausible, realistic-looking videos that are consistent with the training data can produce poor decisions. We address two key questions: (1) when can we trust a world model for planning, and (2) what should we do when we cannot? We introduce CURE (Certify, Upweight, Rollout, Explore), a certificate-guided technique that assesses whether a world model can be trusted for planning and determines what to do next: whether to upweight the training data to debias the model; roll out more samples at test time; or explore by collecting new data where support (how well the training data cover the planner's decisions) is missing. We deploy CURE across multiple domains, spanning simulated physics based on PhyWorld, MuJoCo arm control, and Language-Table pushing in simulation and on real videos. The decisions our certificate trusts have lower regret than those it does not (e.g., 0.06 vs. 0.38 on Language-Table sim and 0.07 vs. 0.38 on MuJoCo contact), whereas video perceptual quality does not identify good decisions. By focusing on decision quality rather than perceptual quality, CURE lowers regret over the best pessimistic decision rule by 10-25% across support levels, and over world-action models, search and policy learning by 41-64%, given the same data.

cs.CV↗

Fast Uncertainty Quantification for Kernel-Based Estimators in Large-Scale Causal Inference

Kernel methods are widely used in causal inference for tasks such as treatment effect estimation, policy evaluation, and policy learning. The bootstrap is a standard tool for uncertainty quantification because of its broad applicability. As increasingly large datasets become available, such as the 2023 U.S. Natality data from the National Vital Statistics System (NVSS), which includes 3,605,081 registered births (of which 2,980,739 form our analytic sample after cohort restrictions), the computational demands of these methods increase substantially. Kernel methods are known to scale poorly with sample size, and this limitation is further exacerbated by the repeated re-fitting required by the bootstrap. As a result, bootstrap-based inference for kernel-based estimators can become computationally infeasible in large-scale settings. In this paper, we address these challenges by extending the causal Bag of Little Bootstraps (cBLB) algorithm of Kosko et al. (2023) to kernel methods. Our approach achieves computational scalability by combining subsampling and resampling while preserving first-order uncertainty quantification and asymptotically correct coverage. We evaluate the method across three representative implementations: kernelized augmented outcome-weighted learning, kernel-based minimax weighting, and double machine learning with kernel support vector machines. We show in simulations that our method yields confidence intervals with nominal coverage at a fraction of the computational cost. We further demonstrate its utility in a real-world application by estimating the effect of any amount of smoking on birth weight, as well as the optimal treatment regime, using the NVSS dataset, where the standard bootstrap is prohibitively expensive computationally and effectively infeasible at this scale.

stat.ME↗

Reinforcement Learning Under Runtime Confounding: Policy Learning and Off-Policy Evaluation

Reinforcement learning (RL) is increasingly used to train robust, adaptable systems across healthcare, e-commerce, and robotics. In these settings, legal, ethical, or logistical concerns may preclude certain key variables that drive decisions from being collected during deployment time. This gives rise to a setting known as runtime confounding: the logged data hold full observability, but deployment time lacks knowledge of certain variables, known as runtime confounders. Unlike single timepoint and non-adaptive settings, runtime confounding under reinforcement learning has the potential to propagate over time, as earlier decisions affect future states, observations, and action assignments, leading to biased learning. We address these problems with three contributions: we (1) develop policy learning algorithms under runtime-confounding, (2) develop semi-parametric efficient off-policy estimation procedures, and (3) provide diagnostics for understanding the price of runtime confounding.

stat.ML↗

More efficient transportation of causal effects using mediators

We develop flexible semiparametric estimators of the average treatment effect transported to a target population that leverage mediators to achieve identification, and prior knowledge of population heterogeneity or summary-level statistics to achieve improved efficiency. We first propose a one-step semiparametric estimator that assumes knowledge of the baseline covariates driving the heterogeneity across populations. This approach improves efficiency by transporting only through these covariates, rather than through the full covariate set. We then introduce a collaborative one-step semiparametric estimator that also improves efficiency without requiring knowledge of which covariates drive study heterogeneity. Collectively, these estimators advance our ability to transport causal effects by improving efficiency in the presence of mediation and intermediate confounding through targeted dimension reduction strategies. We use simulation to examine finite-sample performance and apply our estimators to a large-scale housing mobility trial (MTO) and a hearing intervention trial (ACHIEVE).

stat.ME↗

Reorganizing Graduate Statistical Inference for Modern Biostatistical Practice

First-year graduate courses in statistical inference within PhD programs in biostatistics often follow a classical mathematical statistics progression through probability models, data reduction, point estimation, testing and interval estimation, and asymptotic evaluation. This organization has clear pedagogical strengths, but may provide less guidance when the probability model is not the natural starting point. We propose a complementary target-first organization for a graduate statistical inference course in biostatistics. The course begins with the scientific question and target parameter, then introduces statistical model classes, estimation, first-order sampling behavior, inference, and efficiency within a common framework. Classical methods such as maximum likelihood, method of moments, estimating equations, confidence intervals, hypothesis tests, bootstrap and jackknife methods, and the Cramér--Rao bound are organized within a common inferential workflow. A central advantage of this organization is that the same workflow provides a natural bridge from classical inference to modern statistical problems, including flexible estimation using machine learning and artificial intelligence, policy learning, anytime-valid inference, and modern Bayesian methods.

stat.OT↗

Are Good Generators Good Decision-Makers? Policy Learning for General Interventions via Retargeted Counterfactual Generation

Generative models are increasingly used to support decision-making in complex systems, where interventions may be joint and high-dimensional, and outcomes are high-dimensional. However, using generators for these decision-making settings are challenged by three problems. First, they are often trained on noisy logs with limited intervention data. Second, generator learning can be noisy across different environments. Third, generators are not optimized for the decisions they support. We introduce policy learning via retargeted counterfactual generation, which trains a generator for the decisions it supports in three steps. We (1) learn a doubly-robust, invariant counterfactual generator for high-dimensional interventions and outcomes, (2) conduct policy-learning based on its rollouts, and (3) retarget the generator toward the learned policy's interventions and relearn the policy, so the generator is accurate where decisions are made. Theoretically, our generator's excess counterfactual risk has a doubly robust product remainder, and retargeting removes the worst-case density ratio between the logged and learned policies from the regret. We demonstrate the efficacy of our approach across synthetic data, Cell Painting images of SARS-CoV-2-infected cells, and physical video simulations.

stat.ML↗

Statistical Inference for Continuous Action Bandits

Statistical inference under adaptively collected data is becoming increasingly popular across e-commerce and mobile health. Many methods exist for discrete action settings, ranging from weighting to debiasing approaches. Despite the ubiquity of continuous actions in experimentation from optimal pricing to precision dosing, to the best of our knowledge, methods that support statistical inference under continuous action, adaptively collected data remain underdeveloped. In this work, we extend kernel-smoothed doubly robust estimation from i.i.d. data to adaptively collected data with continuous actions. We study a family of adaptively weighted estimators, establish their mean squared error rates, asymptotic normality, and characterize an estimation lower bound as a function of regret. We conclude with recommendations for experimental design informing efficiency and regret, and support our theory with an extensive simulation study.

stat.ME↗

Efficient transport and generalization of survival treatment effects

Randomized controlled trials provide internally valid estimates of treatment effects, but their results may not directly apply to broader target populations due to differences in baseline covariate distributions, adherence to treatment or variations in outcome mechanisms. Under standard transport and time-to-event identifiability assumptions, we develop nonparametric, debiased machine learning estimators for transporting and generalizing causal survival treatment effect differences from a source population to a target population in discrete time. We derive the efficient influence functions for the transport and generalization survival difference estimands and propose cross-fitted one-step estimators that are doubly robust and achieve semiparametric efficiency bounds under weak regularity conditions. We further introduce estimators that exploit known effect modifier subsets through an additive parameterization of the survival function, reducing the dimensionality of the reweighting and yielding smaller or equal asymptotic variance. We establish asymptotic normality, double robustness, and rates of convergence for all proposed estimators. Finite-sample properties are illustrated through Monte Carlo simulations under flexible and misspecified nuisance estimation scenarios. We apply the methods to data from the Women's Health Initiative to estimate the effect of hormone therapy on coronary heart disease across trial and observational populations.

stat.ME↗

Causal Survival Analysis in Platform Trials with Non-Concurrent Controls

Platform trials allow treatment arms to enter and exit over time while maintaining a shared control arm, yielding concurrent and non-concurrent controls (NCC). Pooling NCC is often motivated as a strategy to improve statistical efficiency, but it is unclear which estimand is targeted, what assumptions justify identification and estimation, and when precision gains are achievable; these questions are further complicated by time-to-event/survival data. Motivated by the Adaptive COVID-19 Treatment Trial (ACTT) platform trial with time to recovery as the primary endpoint, we develop an estimand-first causal survival framework targeting the treatment-specific counterfactual survival curve in the concurrent population and the corresponding functionals including the concurrent restricted mean survival time (RMST). We give nonparametric identification results and formalize conditions that justify pooling using NCC. We study covariate-adjusted outcome-regression (OR) and doubly robust (DR) estimators for the concurrent RMST, comparing concurrent-only versions to pooled-control versions. Pooling improves precision for OR estimators only when the pooling assumption holds and parametric hazard models are correctly specified; otherwise, pooling can induce bias. Moreover, in certain settings, pooling NCC yields no efficiency gain for the DR estimator. Overall, the most robust route to improve precision is to target concurrent causal survival estimands and use a covariate-adjusted DR estimation that uses only concurrent controls. An ACTT application corroborates these results.

stat.ME↗

Data-Adaptive and Model-Robust Covariate Adjustment for Time-to-Event Outcomes in Stratified Randomized Trials

Time-to-event outcomes are commonly used as primary endpoints in randomized clinical trials. Despite this, relatively little work incorporates baseline covariate information while also accounting for stratified randomization, a common form of randomization. Moreover, leveraging efficiency gains using these approaches typically requires pre-specifying a subset of covariates that are most predictive of the outcome -- a challenging task in practice, as most trials collect dozens of potentially prognostic baseline variables. In this work, we build on existing literature to propose a data-adaptive and model-robust covariate adjustment method for time-to-event outcomes. Our approach, based on targeted minimum loss-based estimation, allows for data-adaptive covariate selection and model-robust efficient inference on functionals of the survival curve while accounting for stratification. Through extensive simulations and analysis, we showcase the simplicity and improved precision of our method when the covariate set is not known a priori.

stat.ME↗

Modern Causal Inference Approaches to Improve Power for Subgroup Analysis in Randomized Controlled Trials

Randomized controlled trials (RCTs) often include subgroup analyses to assess whether treatment effects vary across pre-specified patient populations. However, these analyses frequently suffer from small sample sizes which limit the power to detect heterogeneous effects. Power can be improved by leveraging predictors of the outcome -- i.e., through covariate adjustment -- as well as by borrowing external data from similar RCTs or observational studies. The benefits of covariate adjustment may be limited when the trial sample is small. Borrowing external data can increase the effective sample size and improve power, but it introduces two key challenges: (i) integrating data across sources can lead to model misspecification, and (ii) practical violations of the positivity assumption -- where the probability of receiving the target treatment is near-zero for some covariate profiles in the external data -- can lead to extreme inverse-probability weights and unstable inferences, ultimately negating potential power gains. To account for these shortcomings, we present an approach to improving power in pre-planned subgroup analyses of small RCTs that leverages both baseline predictors and external data. We propose debiased estimators that accommodate parametric, machine learning, and nonparametric Bayesian methods. To address practical positivity violations, we introduce three estimators: a covariate-balancing approach, an automated debiased machine learning (DML) estimator, and a calibrated DML estimator. We show improved power in various simulations and offer practical recommendations for the application of the proposed methods. Finally, we apply them to evaluate the effectiveness of citalopram for negative symptoms in first-episode schizophrenia patients across subgroups defined by duration of untreated psychosis, using data from two small RCTs.

stat.ME↗

DeDPO: Debiased Direct Preference Optimization for Diffusion Models

Direct Preference Optimization (DPO) has emerged as a predominant alignment method for diffusion models, facilitating off-policy training without explicit reward modeling. However, its reliance on large-scale, high-quality human preference labels presents a severe cost and scalability bottleneck. To overcome this, We propose a semi-supervised framework augmenting limited human data with a large corpus of unlabeled pairs annotated via cost-effective synthetic AI feedback. Our paper introduces Debiased DPO (DeDPO), which uniquely integrates a debiased estimation technique from causal inference into the DPO objective. By explicitly identifying and correcting the systematic bias and noise inherent in synthetic annotators, DeDPO ensures robust learning from imperfect feedback sources, including self-training and Vision-Language Models (VLMs). Experiments demonstrate that DeDPO is robust to the variations in synthetic labeling methods, achieving performance that matches and occasionally exceeds the theoretical upper bound of models trained on fully human-labeled data. This establishes DeDPO as a scalable solution for human-AI alignment using inexpensive synthetic supervision.

cs.CV↗

Clarifying identification and estimation of treatment effects in the Sequential Parallel Comparison Design

Sequential parallel comparison design (SPCD) clinical trials aim to adjust active treatment effect estimates for placebo response to minimize the impact of placebo responders on the estimates. This is potentially accomplished using a two stage design by measuring treatment effects among all participants during the first stage, then classifying some placebo arm participants as placebo non-responders who will be re-randomized in the second stage. In this paper, we use causal inference tools to clarify under what assumptions treatment effects can be identified in SPCD trials and what effects the conventional estimators target at each stage of the SPCD trial. We further illustrate the highly influential impact of placebo response misclassification on the second stage estimate. We conclude that the conventional SPCD estimators do not target meaningful treatment effects.

stat.ME↗

Identification and estimation of causal effects using non-concurrent controls in platform trials

Platform trials are multi-arm designs that simultaneously evaluate multiple treatments for a single disease within the same overall trial structure. Unlike traditional randomized controlled trials, they allow treatment arms to enter and exit the trial at distinct times while maintaining a control arm throughout. This control arm comprises both concurrent controls, where participants are randomized concurrently to either the treatment or control arm, and non-concurrent controls, who enter the trial when the treatment arm under study is unavailable. While flexible, platform trials introduce the challenge of using non-concurrent controls, raising questions about estimating treatment effects. Specifically, which estimands should be targeted? Under what assumptions can these estimands be identified and estimated? Are there any efficiency gains? In this paper, we discuss issues related to the identification and estimation assumptions of common choices of estimand. We conclude that the most robust strategy to increase efficiency without imposing unwarranted assumptions is to target the concurrent average treatment effect (cATE), the ATE among only concurrent units, using a covariate-adjusted doubly robust estimator. Our studies suggest that, for the purpose of obtaining efficiency gains, collecting important prognostic variables is more important than relying on non-concurrent controls. We also discuss the perils of targeting ATE due to an untestable extrapolation assumption that will often be invalid. We provide simulations illustrating our points and an application to the ACTT platform trial, resulting in a 20% improvement in precision compared to the naive estimator that ignores non-concurrent controls and prognostic variables.

stat.ME↗

Non-parametric efficient estimation of marginal structural models with multi-valued time-varying treatments

Marginal structural models are a popular method for estimating causal effects in the presence of time-varying exposures. In spite of their popularity, no scalable non-parametric estimator exist for marginal structural models with multi-valued and time-varying treatments. In this paper, we use machine learning together with recent developments in semiparametric efficiency theory for longitudinal studies to propose such an estimator. The proposed estimator is based on a study of the non-parametric identifying functional, including first order von-Mises expansions as well as the efficient influence function and the efficiency bound. We show conditions under which the proposed estimator is efficient, asymptotically normal, and sequentially doubly robust in the sense that it is consistent if, for each time point, either the outcome or the treatment mechanism is consistently estimated. We perform a simulation study to illustrate the properties of the estimators, and present the results of our motivating study on a COVID-19 dataset studying the impact of mobility on the cumulative number of observed cases.

stat.ME↗

MambaMixer: Efficient Selective State Space Models with Dual Token and Channel Selection

Recent advances in deep learning have mainly relied on Transformers due to their data dependency and ability to learn at scale. The attention module in these architectures, however, exhibits quadratic time and space in input size, limiting their scalability for long-sequence modeling. Despite recent attempts to design efficient and effective architecture backbone for multi-dimensional data, such as images and multivariate time series, existing models are either data independent, or fail to allow inter- and intra-dimension communication. Recently, State Space Models (SSMs), and more specifically Selective State Space Models, with efficient hardware-aware implementation, have shown promising potential for long sequence modeling. Motivated by the success of SSMs, we present MambaMixer, a new architecture with data-dependent weights that uses a dual selection mechanism across tokens and channels, called Selective Token and Channel Mixer. MambaMixer connects selective mixers using a weighted averaging mechanism, allowing layers to have direct access to early features. As a proof of concept, we design Vision MambaMixer (ViM2) and Time Series MambaMixer (TSM2) architectures based on the MambaMixer block and explore their performance in various vision and time series forecasting tasks. Our results underline the importance of selective mixing across both tokens and channels. In ImageNet classification, object detection, and semantic segmentation tasks, ViM2 achieves competitive performance with well-established vision models and outperforms SSM-based vision models. In time series forecasting, TSM2 achieves outstanding performance compared to state-of-the-art methods while demonstrating significantly improved computational cost. These results show that while Transformers, cross-channel attention, and MLPs are sufficient for good performance in time series forecasting, neither is necessary.

cs.LG↗

Chimera: Effectively Modeling Multivariate Time Series with 2-Dimensional State Space Models

Modeling multivariate time series is a well-established problem with a wide range of applications from healthcare to financial markets. Traditional State Space Models (SSMs) are classical approaches for univariate time series modeling due to their simplicity and expressive power to represent linear dependencies. They, however, have fundamentally limited expressive power to capture non-linear dependencies, are slow in practice, and fail to model the inter-variate information flow. Despite recent attempts to improve the expressive power of SSMs by using deep structured SSMs, the existing methods are either limited to univariate time series, fail to model complex patterns (e.g., seasonal patterns), fail to dynamically model the dependencies of variate and time dimensions, and/or are input-independent. We present Chimera that uses two input-dependent 2-D SSM heads with different discretization processes to learn long-term progression and seasonal patterns. To improve the efficiency of complex 2D recurrence, we present a fast training using a new 2-dimensional parallel selective scan. We further present and discuss 2-dimensional Mamba and Mamba-2 as the spacial cases of our 2D SSM. Our experimental evaluation shows the superior performance of Chimera on extensive and diverse benchmarks, including ECG and speech time series classification, long-term and short-term time series forecasting, and time series anomaly detection.

cs.LG↗

A Stable and Efficient Covariate-Balancing Estimator for Causal Survival Effects

We propose an empirically stable and asymptotically efficient covariate-balancing approach to the problem of estimating survival causal effects in data with conditionally-independent censoring. This addresses a challenge often encountered in state-of-the-art nonparametric methods: the use of inverses of small estimated probabilities and the resulting amplification of estimation error. We validate our theoretical results in experiments on synthetic and semi-synthetic data.

stat.ME↗