arXiv subjects

Sanguo Zhang

Publications and source records attributed to Sanguo Zhang.

11 recordsLinked to original sources

arXiv2026

Chaos Is a LADDER: Domain Generalization Beyond Invariance via Reweighting

Domain generalization (DG) aims to learn from multiple source domains and generalize to unseen target domains. Most DG methods pursue invariance: they seek a causal representation whose prediction rule is invariant across domains. This principle is effective when the causal mechanism is stable, but becomes restrictive when the domain itself modulates how causal content maps to the response. In this case, directly feeding domain style into the predictor can create misleading shortcuts, since style does not by itself cause the response. Yet the apparent chaos of multiple styles can become a ladder: style can locate the unseen target domain among source domains and guide which domain-dependent prediction rules should be trusted. We propose \emph{Latent Adaptive Domain Disentanglement and Environment Reweighting} (LADDER), a fixed-model DG pipeline that learns causal/style representations, freezes the encoders, fits source-specific classifiers, and uses an unlabeled target-domain covariate set only at inference to compute weights over these fixed classifiers, with no target labels or model-state updates. We establish theoretical guarantees for source reweighting and validate LADDER on simulations, FMoW, and a location-grouped iWildCam protocol, with gains in overall and group-averaged accuracy.

stat.ML↗

arXiv2026

Deep Multitask Learning for Mixed-Type Outcomes with Shared Sparsity

Most existing multitask learning approaches are limited by their reliance on task-specific loss functions tailored to the scale and type of each outcome. When outcomes differ across tasks, these losses are generally not directly comparable, which makes it difficult to formulate a unified objective and may limit information sharing across tasks. We propose a multitask transformation framework in which task-specific responses may differ through unknown monotone transformations. Motivated by high-dimensional biological applications in which the predictor dimension may diverge with the sample size while only a common subset of predictors is informative, we consider shared sparsity across tasks. Under this framework, we estimate the target functions and identify important predictors by optimizing a smoothed rank-based criterion with a group-Lasso penalty, implemented through a multitask deep neural network with a shared first layer. We establish the nonasymptotic excess-risk bounds, and variable-selection consistency for the proposed estimator. Simulation studies show that the proposed method achieves competitive prediction and variable-selection performance compared with competing approaches. Analyses of gene-expression studies with continuous, binary, and mixed outcomes further illustrate that the proposed method improves prediction and identifies biologically meaningful shared predictors.

stat.ML↗

arXiv2025

An efficient likelihood-free Bayesian inference method based on sequential neural posterior estimation

Sequential neural posterior estimation (SNPE) techniques have been recently proposed for dealing with simulation-based models with intractable likelihoods. Unlike approximate Bayesian computation, SNPE techniques learn the posterior from sequential simulation using neural network-based conditional density estimators by minimizing a specific loss function. The SNPE method proposed by Lueckmann et al. (2017) used a calibration kernel to boost the sample weights around the observed data, resulting in a concentrated loss function. However, the use of calibration kernels may increase the variances of both the empirical loss and its gradient, making the training inefficient. To improve the stability of SNPE, this paper proposes to use an adaptive calibration kernel and several variance reduction techniques. The proposed method greatly speeds up the process of training and provides a better approximation of the posterior than the original SNPE method and some existing competitors as confirmed by numerical experiments. We also managed to demonstrate the superiority of the proposed method for a high-dimensional model with a real-world dataset.

stat.ML↗

arXiv2024

Deep Transformation Model

There has been a significant recent surge in deep neural network (DNN) techniques. Most of the existing DNN techniques have restricted model formats/assumptions. To overcome their limitations, we propose the nonparametric transformation model, which encompasses many popular models as special cases and hence is less sensitive to model mis-specification. This model also has the potential of accommodating heavy-tailed errors, a robustness property not broadly shared. Accordingly, a new loss function, which fundamentally differs from the existing ones, is developed. For computational feasibility, we further develop a double rectified linear unit (DReLU)-based estimator. To accommodate the scenario with a diverging number of input variables and/or noises, we propose variable selection based on group penalization. We further expand the scope to coherently accommodate censored survival data. The estimation and variable selection properties are rigorously established. Extensive numerical studies, including simulations and data analyses, establish the satisfactory practical utility of the proposed methods.

stat.ME↗

arXiv2023

Simulation-based Bayesian Inference from Privacy Protected Data

Many modern statistical analysis and machine learning applications require training models on sensitive user data. Under a formal definition of privacy protection, differentially private algorithms inject calibrated noise into the confidential data or during the data analysis process to produce privacy-protected datasets or queries. However, restricting access to only privatized data during statistical analysis makes it computationally challenging to make valid statistical inferences. In this work, we propose simulation-based inference methods from privacy-protected datasets. In addition to sequential Monte Carlo approximate Bayesian computation, we adopt neural conditional density estimators as a flexible family of distributions to approximate the posterior distribution of model parameters given the observed private query results. We illustrate our methods on discrete time-series data under an infectious disease model and with ordinary linear regression models. Illustrating the privacy-utility trade-off, our experiments and analysis demonstrate the necessity and feasibility of designing valid statistical inference procedures to correct for biases introduced by the privacy-protection mechanisms.

stat.ML↗

arXiv2022

Biomarker-guided heterogeneity analysis of genetic regulations via multivariate sparse fusion

Heterogeneity is a hallmark of many complex diseases. There are multiple ways of defining heterogeneity, among which the heterogeneity in genetic regulations, for example GEs (gene expressions) by CNVs (copy number variations) and methylation, has been suggested but little investigated. Heterogeneity in genetic regulations can be linked with disease severity, progression, and other traits and is biologically important. However, the analysis can be very challenging with the high dimensionality of both sides of regulation as well as sparse and weak signals. In this article, we consider the scenario where subjects form unknown subgroups, and each subgroup has unique genetic regulation relationships. Further, such heterogeneity is "guided" by a known biomarker. We develop an MSF (Multivariate Sparse Fusion) approach, which innovatively applies the penalized fusion technique to simultaneously determine the number and structure of subgroups and regulation relationships within each subgroup. An effective computational algorithm is developed, and extensive simulations are conducted. The analysis of heterogeneity in the GE-CNV regulations in melanoma and GE-methylation regulations in stomach cancer using the TCGA data leads to interesting findings.

stat.ME↗

arXiv2020

Histopathological imaging features- versus molecular measurements-based cancer prognosis modeling

For most if not all cancers, prognosis is of significant importance, and extensive modeling research has been conducted. With the genetic nature of cancer, in the past two decades, multiple types of molecular data (such as gene expressions and DNA mutations) have been explored. More recently, histopathological imaging data, which is routinely collected in biopsy, has been shown as informative for modeling prognosis. In this study, using the TCGA LUAD and LUSC data as a showcase, we examine and compare modeling lung cancer overall survival using gene expressions versus histopathological imaging features. High-dimensional regularization methods are adopted for estimation and selection. Our analysis shows that gene expressions have slightly better prognostic performance. In addition, most of the gene expressions are found to be weakly correlated imaging features. It is expected that this study can provide some insight into utilizing the two types of important data in cancer prognosis modeling and into lung cancer overall survival.

stat.AP↗

arXiv2016

Identifying Gene-Environment Interactions with A Least Relative Error Approach

For complex diseases, the interactions between genetic and environmental risk factors can have important implications beyond the main effects. Many of the existing interaction analyses conduct marginal analysis and cannot accommodate the joint effects of multiple main effects and interactions. In this study, we conduct joint analysis which can simultaneously accommodate a large number of effects. Significantly different from the existing studies, we adopt loss functions based on relative errors, which offer a useful alternative to the "classic" methods such as the least squares and least absolute deviation. Further to accommodate censoring in the response variable, we adopt a weighted approach. Penalization is used for identification and regularized estimation. Computationally, we develop an effective algorithm which combines the majorize-minimization and coordinate descent. Simulation shows that the proposed approach has satisfactory performance. We also analyze lung cancer prognosis data with gene expression measurements.

stat.ME↗

arXiv2015

A Robust Approach for Identifying Gene-Environment Interactions for Prognosis

For many complex diseases, prognosis is of essential importance. It has been shown that, beyond the main effects of genetic (G) and environmental (E) risk factors, the gene-environment (G$\times$E) interactions also play a critical role. In practice, the prognosis outcome data can be contaminated, and most of the existing methods are not robust to data contamination. In the literature, it has been shown that even a single contaminated observation can lead to severely biased model estimation. In this study, we describe prognosis using an accelerated failure time (AFT) model. An exponential squared loss is proposed to accommodate possible data contamination. A penalization approach is adopted for regularized estimation and marker selection. The proposed method is realized using an effective coordinate descent (CD) and minorization maximization (MM) algorithm. Simulation shows that without contamination, the proposed method has performance comparable to or better than the unrobust alternative. With contamination, it outperforms the unrobust alternative and, under certain scenarios, can be superior to the robust method based on quantile regression. The proposed method is applied to the analysis of TCGA (The Cancer Genome Atlas) lung cancer data. It identifies interactions different from those using the alternatives. The identified marker have important implications and satisfactory stability.

stat.AP↗

arXiv2015

Promoting Similarity of Sparsity Structures in Integrative Analysis with Penalization

For data with high-dimensional covariates but small to moderate sample sizes, the analysis of single datasets often generates unsatisfactory results. The integrative analysis of multiple independent datasets provides an effective way of pooling information and outperforms single-dataset analysis and some alternative multi-datasets approaches including meta-analysis. Under certain scenarios, multiple datasets are expected to share common important covariates, that is, the multiple models have similarity in sparsity structures. However, the existing methods do not have a mechanism to {\it promote} the similarity of sparsity structures in integrative analysis. In this study, we consider penalized variable selection and estimation in integrative analysis. We develop an $L_0$-penalty based approach, which is the first to explicitly promote the similarity of sparsity structures. Computationally it is realized using a coordinate descent algorithm. Theoretically it has the much desired consistency properties. In simulation, it significantly outperforms the competing alternative when the models in multiple datasets share common important covariates. It has better or similar performance as the alternative when the sparsity structures share no similarity. Thus it provides a "safe" choice for data analysis. Applying the proposed method to three lung cancer datasets with gene expression measurements leads to models with significantly more similar sparsity structures and better prediction performance.

stat.ME↗

arXiv2015

Penalized integrative analysis under the accelerated failure time model

For survival data with high-dimensional covariates, results generated in the analysis of a single dataset are often unsatisfactory because of the small sample size. Integrative analysis pools raw data from multiple independent studies with comparable designs, effectively increases sample size, and has better performance than meta-analysis and single-dataset analysis. In this study, we conduct integrative analysis of survival data under the accelerated failure time (AFT) model. The sparsity structures of multiple datasets are described using the homogeneity and heterogeneity models. For variable selection under the homogeneity model, we adopt group penalization approaches. For variable selection under the heterogeneity model, we use composite penalization and sparse group penalization approaches. As a major advancement from the existing studies, the asymptotic selection and estimation properties are rigorously established. Simulation study is conducted to compare different penalization methods and against alternatives. We also analyze four lung cancer prognosis datasets with gene expression measurements.

stat.ME↗