Reference-Based Classification

In reference-based classification, the task is to correctly
predict the label by using serially or spatially diversified samples,
such as evaluation of positive or negative response to drug treatment,
or classification of diseases based on gene microarray responses from
multiple tissues. Variable selection in reference-based multiclass
problems is more challenging than in binary classification. We propose
solving the multiclass support vector machine with a mixed L1/L2 norm
penalty, which enables group variable selection over multiple
references and classes.