Pretrial risk assessment tools are used in jurisdictions across the country to assess the likelihood of "pretrial failure," the event where defendants either fail to appear for court or reoffend. Judicial officers, in turn, use these assessments to determine whether to release or detain defendants during trial. While algorithmic risk assessment tools were designed to predict pretrial failure with greater accuracy relative to judges, there is still concern that both risk assessment recommendations and pretrial decisions are biased against minority groups. In this paper, we develop methods to investigate the association between risk factors and pretrial failure, while simultaneously estimating misclassification rates of pretrial risk assessments and of judicial decisions as a function of defendant race. This approach adds to a growing literature that makes use of outcome misclassification methods to answer questions about fairness in pretrial decision-making. We give a detailed simulation study for our proposed methodology and apply these methods to data from the Virginia Department of Criminal Justice Services. We estimate that the VPRAI algorithm has near-perfect specificity, but its sensitivity differs by defendant race. Judicial decisions also display evidence of bias; we estimate wrongful detention rates of 39.7% and 51.4% among white and Black defendants, respectively.
In generalized regression models the effect of continuous covariates is commonly assumed to be linear. This assumption, however, may be too restrictive in applications and may lead to biased effect estimates and decreased predictive ability. While a multitude of alternatives for the flexible modeling of continuous covariates have been proposed, methods that provide guidance for choosing a suitable functional form are still limited. To address this issue, we propose a detection algorithm that evaluates several approaches for modeling continuous covariates and guides practitioners to choose the most appropriate alternative. The algorithm utilizes a unified framework for tree-structured modeling which makes the results easily interpretable. We assessed the performance of the algorithm by conducting a simulation study. To illustrate the proposed algorithm, we analyzed data of patients suffering from chronic kidney disease.
Muscle volume is a useful quantitative biomarker in sports, but also for the follow-up of degenerative musculo-skelletal diseases. In addition to volume, other shape biomarkers can be extracted by segmenting the muscles of interest from medical images. Manual segmentation is still today the gold standard for such measurements despite being very time-consuming. We propose a method for automatic segmentation of 18 muscles of the lower limb on 3D Magnetic Resonance Images to assist such morphometric analysis. By their nature, the tissue of different muscles is undistinguishable when observed in MR Images. Thus, muscle segmentation algorithms cannot rely on appearance but only on contour cues. However, such contours are hard to detect and their thickness varies across subjects. To cope with the above challenges, we propose a segmentation approach based on a hybrid architecture, combining convolutional and visual transformer blocks. We investigate for the first time the behaviour of such hybrid architectures in the context of muscle segmentation for shape analysis. Considering the consistent anatomical muscle configuration, we rely on transformer blocks to capture the longrange relations between the muscles. To further exploit the anatomical priors, a second contribution of this work consists in adding a regularisation loss based on an adjacency matrix of plausible muscle neighbourhoods estimated from the training data. Our experimental results on a unique database of elite athletes show it is possible to train complex hybrid models from a relatively small database of large volumes, while the anatomical prior regularisation favours better predictions.
Question answering methods are well-known for leveraging data bias, such as the language prior in visual question answering and the position bias in machine reading comprehension (extractive question answering). Current debiasing methods often come at the cost of significant in-distribution performance to achieve favorable out-of-distribution generalizability, while non-debiasing methods sacrifice a considerable amount of out-of-distribution performance in order to obtain high in-distribution performance. Therefore, it is challenging for them to deal with the complicated changing real-world situations. In this paper, we propose a simple yet effective novel loss function with adaptive loose optimization, which seeks to make the best of both worlds for question answering. Our main technical contribution is to reduce the loss adaptively according to the ratio between the previous and current optimization state on mini-batch training data. This loose optimization can be used to prevent non-debiasing methods from overlearning data bias while enabling debiasing methods to maintain slight bias learning. Experiments on the visual question answering datasets, including VQA v2, VQA-CP v1, VQA-CP v2, GQA-OOD, and the extractive question answering dataset SQuAD demonstrate that our approach enables QA methods to obtain state-of-the-art in- and out-of-distribution performance in most cases. The source code has been released publicly in \url{//github.com/reml-group/ALO}.
We systematically study a wide variety of generative models spanning semantically-diverse image datasets to understand and improve the feature extractors and metrics used to evaluate them. Using best practices in psychophysics, we measure human perception of image realism for generated samples by conducting the largest experiment evaluating generative models to date, and find that no existing metric strongly correlates with human evaluations. Comparing to 17 modern metrics for evaluating the overall performance, fidelity, diversity, rarity, and memorization of generative models, we find that the state-of-the-art perceptual realism of diffusion models as judged by humans is not reflected in commonly reported metrics such as FID. This discrepancy is not explained by diversity in generated samples, though one cause is over-reliance on Inception-V3. We address these flaws through a study of alternative self-supervised feature extractors, find that the semantic information encoded by individual networks strongly depends on their training procedure, and show that DINOv2-ViT-L/14 allows for much richer evaluation of generative models. Next, we investigate data memorization, and find that generative models do memorize training examples on simple, smaller datasets like CIFAR10, but not necessarily on more complex datasets like ImageNet. However, our experiments show that current metrics do not properly detect memorization: none in the literature is able to separate memorization from other phenomena such as underfitting or mode shrinkage. To facilitate further development of generative models and their evaluation we release all generated image datasets, human evaluation data, and a modular library to compute 17 common metrics for 9 different encoders at //github.com/layer6ai-labs/dgm-eval.
In survival contexts, substantial literature exists on estimating optimal treatment regimes, where treatments are assigned based on personal characteristics for the purpose of maximizing the survival probability. These methods assume that a set of covariates is sufficient to deconfound the treatment-outcome relationship. Nevertheless, the assumption can be limited in observational studies or randomized trials in which non-adherence occurs. Thus, we propose a novel approach for estimating the optimal treatment regime when certain confounders are not observable and a binary instrumental variable is available. Specifically, via a binary instrumental variable, we propose two semiparametric estimators for the optimal treatment regime by maximizing Kaplan-Meier-like estimators within a pre-defined class of regimes, one of which possesses the desirable property of double robustness. Because the Kaplan-Meier-like estimators are jagged, we incorporate kernel smoothing methods to enhance their performance. Under appropriate regularity conditions, the asymptotic properties are rigorously established. Furthermore, the finite sample performance is assessed through simulation studies. Finally, we exemplify our method using data from the National Cancer Institute's (NCI) prostate, lung, colorectal, and ovarian cancer screening trial.
In an era where scientific experiments can be very costly, multi-fidelity emulators provide a useful tool for cost-efficient predictive scientific computing. For scientific applications, the experimenter is often limited by a tight computational budget, and thus wishes to (i) maximize predictive power of the multi-fidelity emulator via a careful design of experiments, and (ii) ensure this model achieves a desired error tolerance with some notion of confidence. Existing design methods, however, do not jointly tackle objectives (i) and (ii). We propose a novel stacking design approach that addresses both goals. A multi-level reproducing kernel Hilbert space (RKHS) interpolator is first introduced to build the emulator, under which our stacking design provides a sequential approach for designing multi-fidelity runs such that a desired prediction error of $\epsilon > 0$ is met under regularity assumptions. We then prove a novel cost complexity theorem that, under this multi-level interpolator, establishes a bound on the computation cost (for training data simulation) needed to achieve a prediction bound of $\epsilon$. This result provides novel insights on conditions under which the proposed multi-fidelity approach improves upon a conventional RKHS interpolator which relies on a single fidelity level. Finally, we demonstrate the effectiveness of stacking designs in a suite of simulation experiments and an application to finite element analysis.
Scientists continue to develop increasingly complex mechanistic models to reflect their knowledge more realistically. Statistical inference using these models can be challenging since the corresponding likelihood function is often intractable and model simulation may be computationally burdensome. Fortunately, in many of these situations, it is possible to adopt a surrogate model or approximate likelihood function. It may be convenient to conduct Bayesian inference directly with the surrogate, but this can result in bias and poor uncertainty quantification. In this paper we propose a new method for adjusting approximate posterior samples to reduce bias and produce more accurate uncertainty quantification. We do this by optimizing a transform of the approximate posterior that maximizes a scoring rule. Our approach requires only a (fixed) small number of complex model simulations and is numerically stable. We demonstrate good performance of the new method on several examples of increasing complexity.
Threshold selection is a fundamental problem in any threshold-based extreme value analysis. While models are asymptotically motivated, selecting an appropriate threshold for finite samples can be difficult through standard methods. Inference can also be highly sensitive to the choice of threshold. Too low a threshold choice leads to bias in the fit of the extreme value model, while too high a choice leads to unnecessary additional uncertainty in the estimation of model parameters. In this paper, we develop a novel methodology for automated threshold selection that directly tackles this bias-variance trade-off. We also develop a method to account for the uncertainty in this threshold choice and propagate this uncertainty through to high quantile inference. Through a simulation study, we demonstrate the effectiveness of our method for threshold selection and subsequent extreme quantile estimation. We apply our method to the well-known, troublesome example of the River Nidd dataset.
Ensemble forecasts and their combination are explored from the perspective of a probability space. Manipulating ensemble forecasts as discrete probability distributions, multi-model ensembles (MMEs) are reformulated as barycenters of these distributions. Barycenters are defined with respect to a given distance. The barycenter with respect to the L2-distance is shown to be equivalent to the pooling method. Then, the barycenter-based approach is extended to a different distance with interesting properties in the distribution space: the Wasserstein distance. Another interesting feature of the barycenter approach is the possibility to give different weights to the ensembles and so to naturally build weighted MME. As a proof of concept, the L2- and the Wasserstein-barycenters are applied to combine two models from the S2S database, namely the European Centre Medium-Range Weather Forecasts (ECMWF) and the National Centers for Environmental Prediction (NCEP) models. The performance of the two (weighted-) MMEs are evaluated for the prediction of weekly 2m-temperature over Europe for seven winters. The weights given to the models in the barycenters are optimized with respect to two metrics, the CRPS and the proportion of skilful forecasts. These weights have an important impact on the skill of the two barycenter-based MMEs. Although the ECMWF model has an overall better performance than NCEP, the barycenter-ensembles are generally able to outperform both. However, the best MME method, but also the weights, are dependent on the metric. These results constitute a promising first implementation of this methodology before moving to combination of more models.
Difference-in-differences (DID) is a popular approach to identify the causal effects of treatments and policies in the presence of unmeasured confounding. DID identifies the sample average treatment effect in the treated (SATT). However, a goal of such research is often to inform decision-making in target populations outside the treated sample. Transportability methods have been developed to extend inferences from study samples to external target populations; these methods have primarily been developed and applied in settings where identification is based on conditional independence between the treatment and potential outcomes, such as in a randomized trial. This paper develops identification and estimators for effects in a target population, based on DID conducted in a study sample that differs from the target population. We present a range of assumptions under which one may identify causal effects in the target population and employ causal diagrams to illustrate these assumptions. In most realistic settings, results depend critically on the assumption that any unmeasured confounders are not effect measure modifiers on the scale of the effect of interest. We develop several estimators of transported effects, including a doubly robust estimator based on the efficient influence function. Simulation results support theoretical properties of the proposed estimators. We discuss the potential application of our approach to a study of the effects of a US federal smoke-free housing policy, where the original study was conducted in New York City alone and the goal is extend inferences to other US cities.