Estimating relevant labels Jointly based on multivariate regression

Recent advances have shown that some intrinsic traits of personality are reflected in the resting-state functional brain network (FBN), which opens a new way of evaluating personality more objectively (relative to the traditional questionnaire-based methods). However, current studies evaluate different factors of personality independently without considering their potential relationship. To address this problem, we propose to estimate several factor scores jointly for personality evaluation by introducing a factor-similarity matrix that can be automatically updated according to the FBN data. Additionally, an L
₁
-norm regularizer is used in the estimating method to obtain a sparse solution toward a better interpretability. As a result, we can realize multi-variate linear regression, feature selection and factor-relationship learning in a unified framework. Finally, we design an alternating optimization algorithm to solve the proposed method. Experimental results on the Human Connectome Project (HCP) dataset show that the proposed method, with the automatically learned factor relationship, can achieve a higher estimated accuracy than the related method.