Fit multinomial logistic regression with a penalty on the nuclear norm of the estimated regression coefficient matrix, using proximal gradient descent. Reference: Powers S, Hastie T and Tibshirani R (2018) <doi:10.1177/1471082X18777669>.
