The Bayesian Elastic Net: Classifying Multi-Task Gene-Expression Data

摘要: Highly correlated relevant features are frequently encountered in variable-selection problems, with gene-expression analysis an important example. It is desirable to select all of these highly simultaneously as a group, for better model interpretation and robustness. Further, irrelevant should be excluded, resulting sparse solution (of importance avoiding over-fitting limited data). We address the problem grouped variable selection by introducing new Bayesian Elastic Net model. One advantage proposed that imposing priors on individual parameters Laplace distribution, we reduce number tuning one, compared two such original Net. In addition, extend probit regression, order deal classification problems but set covariates (features). Extension multi-task learning also considered, inference performed using variational analysis. The validated first performing experiments simulated data previously published data; perform comparisons Lasso. Finally, present analyze time-evolving properties influenza, measured blood samples from human subjects recent challenge study.