A solution to the learning dilemma for recurrent networks of spiking neurons

I just saw Mass present this paper at HBP summit and came to post it here. It sounds fairly interesting since it can perform gradients without reverse mode differentiation.