Multi-task learning is increasingly used to investigate the association structure between multiple responses and a single set of predictor variables in many applications. In the era of big data, the coexistence of incomplete outcomes, large number of responses, and high dimensionality in predictors poses unprecedented challenges in estimation, prediction, and computation. In this paper, we propose a scalable and computationally efficient procedure, called PEER, for large-scale multi-response regression with incomplete outcomes, where both the numbers of responses and predictors can be high-dimensional. Motivated by sparse factor regression, we convert the multi-response regression into a set of univariate-response regressions, which can be efficiently implemented in parallel. Under some mild regularity conditions, we show that PEER enjoys nice sampling properties including consistency in estimation, prediction, and variable selection. Extensive simulation studies show that our proposal compares favorably with several existing methods in estimation accuracy, variable selection, and computation efficiency.
The missing data issue is ubiquitous in health studies. Variable selection in the presence of both missing covariates and outcomes is an important statistical research topic but has been less studied. Existing literature focuses on parametric regression techniques that provide direct parameter estimates of the regression model. Flexible nonparametric machine learning methods considerably mitigate the reliance on the parametric assumptions, but do not provide as naturally defined variable importance measure as the covariate effect native to parametric models. We investigate a general variable selection approach when both the covariates and outcomes can be missing at random and have general missing data patterns. This approach exploits the flexibility of machine learning modeling techniques and bootstrap imputation, which is amenable to nonparametric methods in which the covariate effects are not directly available. We conduct expansive simulations investigating the practical operating characteristics of the proposed variable selection approach, when combined with four tree-based machine learning methods, XGBoost, Random Forests, Bayesian Additive Regression Trees (BART) and Conditional Random Forests, and two commonly used parametric methods, lasso and backward stepwise selection. Numeric results suggest that when combined with bootstrap imputation, XGBoost and BART have the overall best variable selection performance with respect to the $F_1$ score and Type I error across various settings. In general, there is no significant difference in the variable selection performance due to imputation methods. We further demonstrate the methods via a case study of risk factors for 3-year incidence of metabolic syndrome with data from the Study of Women's Health Across the Nation.
Explanatory variables in a predictive regression typically exhibit low signal strength and various degrees of persistence. Variable selection in such a context is of great importance. In this paper, we explore the pitfalls and possibilities of the LASSO methods in this predictive regression framework. In the presence of stationary, local unit root, and cointegrated predictors, we show that the adaptive LASSO cannot asymptotically eliminate all cointegrating variables with zero regression coefficients. This new finding motivates a novel post-selection adaptive LASSO, which we call the twin adaptive LASSO (TAlasso), to restore variable selection consistency. Accommodating the system of heterogeneous regressors, TAlasso achieves the well-known oracle property. In contrast, conventional LASSO fails to attain coefficient estimation consistency and variable screening in all components simultaneously. We apply these LASSO methods to evaluate the short- and long-horizon predictability of S\&P 500 excess returns.
When fitting statistical models, some predictors are often found to be correlated with each other, and functioning together. Many group variable selection methods are developed to select the groups of predictors that are closely related to the continuous or categorical response. These existing methods usually assume the group structures are well known. For example, variables with similar practical meaning, or dummy variables created by categorical data. However, in practice, it is impractical to know the exact group structure, especially when the variable dimensional is large. As a result, the group variable selection results may be selected. To solve the challenge, we propose a two-stage approach that combines a variable clustering stage and a group variable stage for the group variable selection problem. The variable clustering stage uses information from the data to find a group structure, which improves the performance of the existing group variable selection methods. For ultrahigh dimensional data, where the predictors are much larger than observations, we incorporated a variable screening method in the first stage and shows the advantages of such an approach. In this article, we compared and discussed the performance of four existing group variable selection methods under different simulation models, with and without the variable clustering stage. The two-stage method shows a better performance, in terms of the prediction accuracy, as well as in the accuracy to select active predictors. An athlete's data is also used to show the advantages of the proposed method.
Various unsupervised greedy selection methods have been proposed as computationally tractable approximations to the NP-hard subset selection problem. These methods rely on sequentially selecting the variables that best improve performance with respect to a selection criterion. Theoretical results exist that provide performance bounds and enable "lazy greedy" efficient implementations for selection criteria that satisfy a diminishing returns property known as submodularity. This has motivated the development of variable selection algorithms based on mutual information and frame potential. Recently, the authors introduced Forward Selection Component Analysis (FSCA) which uses variance explained as its selection criterion. While this criterion is not submodular, FSCA has been shown to be highly effective for applications such as measurement plan optimisation. In this paper a "lazy" implementation of the FSCA algorithm (L-FSCA) is proposed, which, although not equivalent to FSCA due to the absence of submodularity, has the potential to yield comparable performance while being up to an order of magnitude faster to compute. The efficacy of L-FSCA is demonstrated by performing a systematic comparison with FSCA and five other unsupervised variable selection methods from the literature using simulated and real-world case studies. Experimental results confirm that L-FSCA yields almost identical performance to FSCA while reducing computation time by between 22% and 94% for the case studies considered.
Variable importance measures are the main tools to analyze the black-box mechanisms of random forests. Although the mean decrease accuracy (MDA) is widely accepted as the most efficient variable importance measure for random forests, little is known about its statistical properties. In fact, the definition of MDA varies across the main random forest software. In this article, our objective is to rigorously analyze the behavior of the main MDA implementations. Consequently, we mathematically formalize the various implemented MDA algorithms, and then establish their limits when the sample size increases. This asymptotic analysis reveals that these MDA versions differ as importance measures, since they converge towards different quantities. More importantly, we break down these limits into three components: the first two terms are related to Sobol indices, which are well-defined measures of a covariate contribution to the response variance, widely used in the sensitivity analysis field, as opposed to the third term, whose value increases with dependence within covariates. Thus, we theoretically demonstrate that the MDA does not target the right quantity to detect influential covariates in a dependent setting, a fact that has already been noticed experimentally. To address this issue, we define a new importance measure for random forests, the Sobol-MDA, which fixes the flaws of the original MDA, and consistently estimates the accuracy decrease of the forest retrained without a given covariate, but with an efficient computational cost. The Sobol-MDA empirically outperforms its competitors on both simulated and real data for variable selection. An open source implementation in R and C++ is available online.
Deep learning has been the engine powering many successes of data science. However, the deep neural network (DNN), as the basic model of deep learning, is often excessively over-parameterized, causing many difficulties in training, prediction and interpretation. We propose a frequentist-like method for learning sparse DNNs and justify its consistency under the Bayesian framework: the proposed method could learn a sparse DNN with at most $O(n/\log(n))$ connections and nice theoretical guarantees such as posterior consistency, variable selection consistency and asymptotically optimal generalization bounds. In particular, we establish posterior consistency for the sparse DNN with a mixture Gaussian prior, show that the structure of the sparse DNN can be consistently determined using a Laplace approximation-based marginal posterior inclusion probability approach, and use Bayesian evidence to elicit sparse DNNs learned by an optimization method such as stochastic gradient descent in multiple runs with different initializations. The proposed method is computationally more efficient than standard Bayesian methods for large-scale sparse DNNs. The numerical results indicate that the proposed method can perform very well for large-scale network compression and high-dimensional nonlinear variable selection, both advancing interpretable machine learning.
Recent evidence shows that deep learning models trained on electronic health records from millions of patients can deliver substantially more accurate predictions of risk compared to their statistical counterparts. While this provides an important opportunity for improving clinical decision-making, the lack of interpretability is a major barrier to the incorporation of these black-box models in routine care, limiting their trustworthiness and preventing further hypothesis-testing investigations. In this study, we propose two methods, namely, model distillation and variable selection, to untangle hidden patterns learned by an established deep learning model (BEHRT) for risk association identification. Due to the clinical importance and diversity of heart failure as a phenotype, it was used to showcase the merits of the proposed methods. A cohort with 788,880 (8.3% incident heart failure) patients was considered for the study. Model distillation identified 598 and 379 diseases that were associated and dissociated with heart failure at the population level, respectively. While the associations were broadly consistent with prior knowledge, our method also highlighted several less appreciated links that are worth further investigation. In addition to these important population-level insights, we developed an approach to individual-level interpretation to take account of varying manifestation of heart failure in clinical practice. This was achieved through variable selection by detecting a minimal set of encounters that can maximally preserve the accuracy of prediction for individuals. Our proposed work provides a discovery-enabling tool to identify risk factors in both population and individual levels from a data-driven perspective. This helps to generate new hypotheses and guides further investigations on causal links.
This work considers variational Bayesian inference as an inexpensive and scalable alternative to a fully Bayesian approach in the context of sparsity-promoting priors. In particular, the priors considered arise from scale mixtures of Normal distributions with a generalized inverse Gaussian mixing distribution. This includes the variational Bayesian LASSO as an inexpensive and scalable alternative to the Bayesian LASSO introduced in [65]. It also includes a family of priors which more strongly promote sparsity. For linear models the method requires only the iterative solution of deterministic least squares problems. Furthermore, for p unknown covariates the method can be implemented exactly online with a cost of $O(p^3)$ in computation and $O(p^2)$ in memory per iteration -- in other words, the cost per iteration is independent of n, and in principle infinite data can be considered. For large $p$ an approximation is able to achieve promising results for a cost of $O(p)$ per iteration, in both computation and memory. Strategies for hyper-parameter tuning are also considered. The method is implemented for real and simulated data. It is shown that the performance in terms of variable selection and uncertainty quantification of the variational Bayesian LASSO can be comparable to the Bayesian LASSO for problems which are tractable with that method, and for a fraction of the cost. The present method comfortably handles $n = 65536$, $p = 131073$ on a laptop in less than 30 minutes, and $n = 10^5$, $p = 2.1 \times 10^6$ overnight.
This paper develops a sparsity-inducing version of Bayesian Causal Forests, a recently proposed nonparametric causal regression model that employs Bayesian Additive Regression Trees and is specifically designed to estimate heterogeneous treatment effects using observational data. The sparsity-inducing component we introduce is motivated by empirical studies where not all the available covariates are relevant, leading to different degrees of sparsity underlying the surfaces of interest in the estimation of individual treatment effects. The extended version presented in this work, which we name Shrinkage Bayesian Causal Forest, is equipped with an additional pair of priors allowing the model to adjust the weight of each covariate through the corresponding number of splits in the tree ensemble. These priors improve the model's adaptability to sparse data generating processes and allow to perform fully Bayesian feature shrinkage in a framework for treatment effects estimation, and thus to uncover the moderating factors driving heterogeneity. In addition, the method allows prior knowledge about the relevant confounding covariates and the relative magnitude of their impact on the outcome to be incorporated in the model. We illustrate the performance of our method in simulated studies, in comparison to Bayesian Causal Forest and other state-of-the-art models, to demonstrate how it scales up with an increasing number of covariates and how it handles strongly confounded scenarios. Finally, we also provide an example of application using real-world data.
In supervised classification problems, the test set may contain data points belonging to classes not observed in the learning phase. Moreover, the same units in the test data may be measured on a set of additional variables recorded at a subsequent stage with respect to when the learning sample was collected. In this situation, the classifier built in the learning phase needs to adapt to handle potential unknown classes and the extra dimensions. We introduce a model-based discriminant approach, Dimension-Adaptive Mixture Discriminant Analysis (D-AMDA), which can detect unobserved classes and adapt to the increasing dimensionality. Model estimation is carried out via a full inductive approach based on an EM algorithm. The method is then embedded in a more general framework for adaptive variable selection and classification suitable for data of large dimensions. A simulation study and an artificial experiment related to classification of adulterated honey samples are used to validate the ability of the proposed framework to deal with complex situations.
Human interpretability of deep neural networks' decisions is crucial, especially in domains where these directly affect human lives. Counterfactual explanations of already trained neural networks can be generated by perturbing input features and attributing importance according to the change in the classifier's outcome after perturbation. Perturbation can be done by replacing features using heuristic or generative in-filling methods. The choice of in-filling function significantly impacts the number of artifacts, i.e., false-positive attributions. Heuristic methods result in false-positive artifacts because the image after the perturbation is far from the original data distribution. Generative in-filling methods reduce artifacts by producing in-filling values that respect the original data distribution. However, current generative in-filling methods may also increase false-negatives due to the high correlation of in-filling values with the original data. In this paper, we propose to alleviate this by generating in-fillings with the statistically-grounded Knockoffs framework, which was developed by Barber and Cand\`es in 2015 as a tool for variable selection with controllable false discovery rate. Knockoffs are statistically null-variables as decorrelated as possible from the original data, which can be swapped with the originals without changing the underlying data distribution. A comparison of different in-filling methods indicates that in-filling with knockoffs can reveal explanations in a more causal sense while still maintaining the compactness of the explanations.
In this paper we propose the adaptive lasso for predictive quantile regression (ALQR). Reflecting empirical findings, we allow predictors to have various degrees of persistence and exhibit different signal strengths. The number of predictors is allowed to grow with the sample size. We study regularity conditions under which stationary, local unit root, and cointegrated predictors are present simultaneously. We next show the convergence rates, model selection consistency, and asymptotic distributions of ALQR. We apply the proposed method to the out-of-sample quantile prediction problem of stock returns and find that it outperforms the existing alternatives. We also provide numerical evidence from additional Monte Carlo experiments, supporting the theoretical results.
Mendelian randomization is the use of genetic variants to assess the existence of a causal relationship between a risk factor and an outcome of interest. Here, we focus on two-sample summary-data Mendelian randomization analyses with many correlated variants from a single gene region, and particularly on cis-Mendelian randomization studies which use protein expression as a risk factor. Such studies must rely on a small, curated set of variants from the studied region; using all variants in the region requires inverting an ill-conditioned genetic correlation matrix and results in numerically unstable causal effect estimates. We review methods for variable selection and estimation in cis-Mendelian randomization with summary-level data, ranging from stepwise pruning and conditional analysis to principal components analysis, factor analysis and Bayesian variable selection. In a simulation study, we show that the various methods have a comparable performance in analyses with large sample sizes and strong genetic instruments. However, when weak instrument bias is suspected, factor analysis and Bayesian variable selection produce more reliable inferences than simple pruning approaches, which are often used in practice. We conclude by examining two case studies, assessing the effects of LDL-cholesterol and serum testosterone on coronary heart disease risk using variants in the HMGCR and SHBG gene regions respectively.
The purpose of this Conference is to present the main lines of base projects that are founded on research already begun in previous years. In this sense, this manuscript will present the main lines of research in Diabetes Mellitus type 1 and Machine Learning techniques in an Internet of Things environment, so that we can summarize the future lines to be developed as follows: data collection through biosensors, massive data processing in the cloud, interconnection of biodevices, local computing vs. cloud computing, and possibilities of machine learning techniques to predict blood glucose values, including both variable selection algorithms and predictive techniques.
Mixed Integer Programming (MIP) solvers rely on an array of sophisticated heuristics developed with decades of research to solve large-scale MIP instances encountered in practice. Machine learning offers to automatically construct better heuristics from data by exploiting shared structure among instances in the data. This paper applies learning to the two key sub-tasks of a MIP solver, generating a high-quality joint variable assignment, and bounding the gap in objective value between that assignment and an optimal one. Our approach constructs two corresponding neural network-based components, Neural Diving and Neural Branching, to use in a base MIP solver such as SCIP. Neural Diving learns a deep neural network to generate multiple partial assignments for its integer variables, and the resulting smaller MIPs for un-assigned variables are solved with SCIP to construct high quality joint assignments. Neural Branching learns a deep neural network to make variable selection decisions in branch-and-bound to bound the objective value gap with a small tree. This is done by imitating a new variant of Full Strong Branching we propose that scales to large instances using GPUs. We evaluate our approach on six diverse real-world datasets, including two Google production datasets and MIPLIB, by training separate neural networks on each. Most instances in all the datasets combined have $10^3-10^6$ variables and constraints after presolve, which is significantly larger than previous learning approaches. Comparing solvers with respect to primal-dual gap averaged over a held-out set of instances, the learning-augmented SCIP is 2x to 10x better on all datasets except one on which it is $10^5$x better, at large time limits. To the best of our knowledge, ours is the first learning approach to demonstrate such large improvements over SCIP on both large-scale real-world application datasets and MIPLIB.
Reducing dimensionality is a key preprocessing step in many data analysis applications to address the negative effects of the curse of dimensionality and collinearity on model performance and computational complexity, to denoise the data or to reduce storage requirements. Moreover, in many applications it is desirable to reduce the input dimensions by choosing a subset of variables that best represents the entire set without any a priori information available. Unsupervised variable selection techniques provide a solution to this second problem. An autoencoder, if properly regularized, can solve both unsupervised dimensionality reduction and variable selection, but the training of large neural networks can be prohibitive in time sensitive applications. We present an approach called Recovery of Linear Components (RLC), which serves as a middle ground between linear and non-linear dimensionality reduction techniques, reducing autoencoder training times while enhancing performance over purely linear techniques. With the aid of synthetic and real world case studies, we show that the RLC, when compared with an autoencoder of similar complexity, shows higher accuracy, similar robustness to overfitting, and faster training times. Additionally, at the cost of a relatively small increase in computational complexity, RLC is shown to outperform the current state-of-the-art for a semiconductor manufacturing wafer measurement site optimization application.
AEGIS study possesses unique information on longitudinal changes in circulating glucose through continuous glucose monitoring technology (CGM). However, as usual in longitudinal medical studies, there is a significant amount of missing data in the outcome variables. For example, 40 percent of glycosylated hemoglobin (A1C) biomarker data are missing five years ahead. With the purpose to reduce the impact of this issue, this article proposes a new data analysis framework based on learning in reproducing kernel Hilbert spaces (RKHS) with missing responses that allows to capture non-linear relations between variable studies in different supervised modeling tasks. First, we extend the Hilbert-Schmidt dependence measure to test statistical independence in this context introducing a new bootstrap procedure, for which we prove consistency. Next, we adapt or use existing models of variable selection, regression, and conformal inference to obtain new clinical findings about glucose changes five years ahead with the AEGIS data. The most relevant findings are summarized below: i) We identify new factors associated with long-term glucose evolution; ii) We show the clinical sensibility of CGM data to detect changes in glucose metabolism; iii) We can improve clinical interventions based on our algorithms' expected glucose changes according to patients' baseline characteristics.
In this paper we propose a regularization approach for network modeling of German power derivative market. To deal with the large portfolio, we combine high-dimensional variable selection techniques with dynamic network analysis. The estimated sparse interconnectedness of the full German power derivative market, clearly identify the significant channels of relevant potential risk spillovers. Our empirical findings show the importance of interdependence between different contract types, and identify the main risk contributors. We further observe strong pairwise interconnections between the neighboring contracts especially for the spot contracts trading in the peak hours, its implications for regulators and investors are also discussed. The network analysis of the full German power derivative market helps us to complement a full picture of system risk, and have a better understanding of the German power market functioning and environment.
We introduce tools for controlled variable selection to economists. In particular, we apply a recently introduced aggregation scheme for false discovery rate (FDR) control to German administrative data to determine the parts of the individual employment histories that are relevant for the career outcomes of women. Our results suggest that career outcomes can be predicted based on a small set of variables, such as daily earnings, wage increases in combination with a high level of education, employment status, and working experience.