Proving that safety-critical neural networks do what they're supposed to: where we are, where we're going (part 1 of 2) - composition.al
deciding what properties to verify about a system is hard in the the same way that security threat modeling is hard. You can’t just ask, “Is this piece of software secure?” You can’t just ask, “Is this neural network safe to use?”
but Hawking is not impressed.
Go to the
to explore the network more!