This is a very interesting paper. Question though: it seems the cells gates since they're updated using a "global" gradient descent that it isn't truly parallel

This is a very interesting paper. Question though: it seems the cells gates since they're updated using a "global" gradient descent that it isn't truly parallel. Is there any promise towards a strictly local weight adjustment method ?