Isn't "replication training" just adversarial training, as in GANs? One network tries to clean-room implement the hash function. The other network tries to fi

Isn't "replication training" just adversarial training, as in GANs? One network tries to clean-room implement the hash function. The other network tries to find an input for which the reference implementation behaves differently.