In principle, can brms handle Gaussian Processes for a categorical (~multinomial) model? This would presumably apply the correlation matrix to each of the sub-functions within the categorical model. Is the estimation likely to be unproblematic?
It can handle it in principle. how well estimation will work is something you have to try out yourself.
Out of curiosity, I’m wondering about correlations across the matrices. That is, brms uses the
(1 | x | ID) syntax to for correlated random effects. Out of curiosity, what would brms do with
(1 | x | gr(phylo, cov = A)) in a categorical model?