With the advancement of modern scientific research, multimodal data is increasingly being collected from multiple sources or types. For outcomes derived from generalized linear models with high-dimensional and multimodal covariates, we develop two distinct factor-adjusted tests to assess the significance of high-dimensional modality data and specific low-dimensional linear combinations of predictors from one or more modalities, respectively. First, we propose a factor-adjusted decorrelated score test to evaluate the significance of a single modality. This approach simultaneously transforms a high-dimensional test into a fixed low-dimensional one while addressing the impact of high-dimensional nuisance parameters. Second, we construct a factor-adjusted Wald test based on partial penalized estimation to assess the significance of certain low-dimensional combinations of variables from one or more modalities. The limiting distributions of these two proposed tests are analyzed under both the null hypothesis and local alternatives to characterize the asymptotic type-I errors and powers. The finite sample performance of our proposed tests is evaluated through simulations and further demonstrated with a breast cancer dataset.

In this paper, we consider parameter estimation, kink points testing and statistical inference for a longitudinal multi-kink expectile regression model with nonignorable dropout. In order to accommodate both within-subject correlations and nonignorable dropout, the bias-corrected generalized estimating equations are constructed by combining the inverse probability weighting and quadratic inference function approaches. The estimators for the kink locations and regression coefficients are obtained by using the generalized method of moments. A selection procedure based on a modified BIC is applied to estimate the number of kink points. We theoretically demonstrate the number selection consistency of kink points and the asymptotic normality of all estimators. A weighted cumulative sum type statistic is proposed to test the existence of kink effects at a given expectile, and its limiting distributions are derived under both the null and the local alternative hypotheses. Simulation studies show that the proposed estimators and test have desirable finite sample performance in both homoscedastic and heteroscedastic errors. An application to the Nation Growth, Lung and Health Study dataset is also presented.

Social media analytics have played an important role in disaster identification. Recent advances in deep learning (DL) technologies have been applied to design disaster classification models. However, the DL-based models are hindered by insufficient training samples, because data collection and labeling are very expensive and time-consuming. To solve this issue, a privacy-preserving federated transfer learning approach for disaster classification (FedTL) is proposed, which can allow distributed social computing nodes to collaboratively train a comprehensive model. In the FedTL, Paillier homomorphic encryption method is used to protect the social computing nodes' data privacy. In particular, the transfer learning technology is adopted as a novel application to reduce the computation and communication costs in the federated learning system. The FedTL is verified by a real disaster image dataset collected from social networks. Theoretical analyses and experiment results show that the FedTL is effective, secure, efficient. In addition, the FedTL is highly extensible and can be easily applied in other transfer learning models.

In this article, we investigate parameter estimation, kink points testing and statistical inference for a longitudinal multikink expectile regression model. The estimators for the kink locations and regression coefficients are obtained by using a bootstrap restarting iterative algorithm to avoid local minima. A backward selection procedure based on a modified BIC is applied to estimate the number of kink points. We theoretically demonstrate the number selection consistency of kink points and the asymptotic normality of all estimators. In particular, the estimators of kink locations are shown to achieve root-n consistency. A weighted cumulative sum type statistic is proposed to test the existence of kink effects at a given expectile, and its limiting distributions are derived under both the null and the local alternative hypotheses. The traditional Wald-type and cluster bootstrap confidence intervals for kink locations are also constructed. Simulation studies show that the proposed estimators and test have desirable finite sample performance in both homoscedastic and heteroscedastic errors. Two applications to the Nation Growth, Lung and Health Study and Capital Bike sharing dataset in Washington D.C. are also presented. The R codes for simulation studies and the real data are available at https://github.com/wangleink/MKER.

The kth ( ) power expectile regression (ER) can balance robustness and effectiveness between the ordinary quantile regression and ER simultaneously. Motivated by a longitudinal ACTG 193A data with nonignorable dropouts, we propose a two-stage estimation procedure and statistical inference methods based on the kth power ER and empirical likelihood to accommodate both the within-subject correlations and nonignorable dropouts. Firstly, we construct the bias-corrected generalized estimating equations by combining the kth power ER and inverse probability weighting approaches. Subsequently, the generalized method of moments is utilized to estimate the parameters in the nonignorable dropout propensity based on sufficient instrumental estimating equations. Secondly, in order to incorporate the within-subject correlations under an informative working correlation structure, we borrow the idea of quadratic inference function to obtain the improved empirical likelihood procedures. The asymptotic properties of the corresponding estimators and their confidence regions are derived. The finite-sample performance of the proposed estimators is studied through simulation and an application to the ACTG 193A data is also presented.