Learning gradient-based ICA by neurally estimating mutual information

DRL
Re-assign community