Identification via compressed data

Deception With Side Information in Biometric Authentication Systems

“…In this paper, we are interested in the minimal achievable deception exponent, which is the best the adversary can do. Based on the minimal achievable deception exponent, the designer of the biometric authentication system can choose an appropriate ∆ value that on one hand, limits the probability of successful deception below the tolerance level, and on the other hand, does not cause too large a probability of false rejection when the legitimate user is authenticated [1], [2].…”

“…In a biometric authentication system, successful deception happens when an adversary impersonates a legitimate user by faking a biometric feature close enough to the enrollment and then deceives the authentication system. The first study on the deception probability in the biometric authentication system is [1], where the authors studied the deception in an authentication system where the enrollment is compressed. The authors obtained the optimal trade-off between the compression rate and the exponent of the probability of successful deception when the adversary has no side information.…”

Compression for Quadratic Similarity Queries

“…according to a given distribution. The setting is closely related to the problem considered by Ahlswede et al [1], where the focus was only on discrete sources. In [1], the authors attempt to attack the more general problem where both false positives and false negatives are allowed.…”

“…The setting is closely related to the problem considered by Ahlswede et al [1], where the focus was only on discrete sources. In [1], the authors attempt to attack the more general problem where both false positives and false negatives are allowed. In this general case, it was demonstrated in [1] that the question of 'achievable rate' is uninteresting and only the error exponent is studied (in the current paper, where false negatives are not allowed, we show that the rate question becomes interesting again).…”

“…In [1], the authors attempt to attack the more general problem where both false positives and false negatives are allowed. In this general case, it was demonstrated in [1] that the question of 'achievable rate' is uninteresting and only the error exponent is studied (in the current paper, where false negatives are not allowed, we show that the rate question becomes interesting again). We should also note that the error exponent results in [1] are parameterized by an auxiliary random variable with unbounded alphabet cardinality, rendering those quantities incomputable, and therefore of limited practical interest.…”