Then we have the face recognition problem, where we need to perform face verification, but rather for a group of people.

In other words identifying whether there is a known person in a group of people:

Although face recognition and verification can be thought as the same problem, the reason we treat them as different is because the face recognition is much harder.

For instance, if we achieve the face verification of 98% accuracy, if a person is the one that it claims, it may be not that bad, so maybe this accuracy could be, in some cases, acceptable. Now, if we apply that model with 2% error rate to the face recognition, with 16 people in this instance, then we have a 32% error rate, since we have 2% for each of these photos:

And obviously, it is not going to work well, since this level of error is quite significant. So, for face recognition to work well and have a reasonable level of accuracy, also taking into account the sensitive information of the topic, we'll need something like 99.999% accuracy.