Comment by strbean

Would these alternatives to backprop make it more feasible to have constant live-training going on in a model? Giving it something akin to neuro-plasticity?