Very neat stuff! So one question is, if we had an analog computer that could run these flows exactly, would we get better results if we ran the gradient flow or this central flow?