Comment by byteware

although zero knowledge proofs can compress the "proof of gradient" so it takes much less time to verify than recompute, it take orders of more work to generate the proof, moreover can we merge multiple "improvements"? if not then upon the selection of the "winner" all those are lost?