Skip to main content
. Author manuscript; available in PMC: 2021 Jan 1.
Published in final edited form as: Proc IEEE Inst Electr Electron Eng. 2019 Oct 29;108(1):125–162. doi: 10.1109/JPROC.2019.2947272

Fig. 9.

Fig. 9.

(a) The JCRMML framework [193] performs joint classification and regression via multimodal multitask learning to identify disease-sensitive and cognition-relevant biomarkers from brain imaging genomic data. The identified biomarkers could predict not only disease status, but also cognitive functions to help us better understand the underlying mechanism from gene to brain structure and function, and to cognition and disease. (b) Illustration of the JCRMML feature weight matrix WT. The group l1-norm (G1-norm) learns the group-wise weights for features within a single modality for each task (i.e., outcome) and the l2,1-norm selects features associated with most tasks. [Images are reproduced here with permission from Oxford University Press [193]].