The article suggests they really didn't make any ml based approach where they take in huge numbers of case mates to train a system, so I think it can't be used

The article suggests they really didn't make any ml based approach where they take in huge numbers of case mates to train a system, so I think it can't be used to say anything about that f that approach works.