approach and test the ability of our framework to discrimi
nate between the previously unseen brands. For each itera
tion, we held out a different pairing of two brands (e.g., Dis
ney and Gucci) from the training set, and the model was
trained using the remaining 42 brands (Figure 2, Panel A).
Specifically, training involved regressing the activation
level of each voxel on the set of personality features of the
training brands obtained from the factor analysis (Figure 2,
Panel B). We used the derived maximum likelihood esti
mates as terms, which we then combined with the personal
ity factor scores of each holdout brand to form a predicted
fMRI image. We iterated this leave-two-out train-test proce
dure 946 times to hold out each of the possible brand pairs
(Figure 2, Panel C).