An adaptive automated method for identity verification with performance guarantees

Abstract

Businesses often rely on customers that access an automated resource to receive information, service or products. We present an adaptive method for automated authentication with performance guarantee of probabilistic error bounds, referred to as identity verification. A database includes a record of identifiers for each identity. The identifiers are partitioned into groups, where identifiers in the same group are correlated while identifiers in different groups are independent. A claimant requesting access into the system is probed with a sequence of identifiers. The response to an identifier can be a match, no-match, or ambiguous. Each identifier is characterized by prior response probabilities for a legitimate claimant and for an impostor. Impostor’s response probabilities are updated during a session. The method guarantees that the probabilities of accepting an impostor or rejecting a legitimate claimant do not exceed specified parameters. Once a given number of identifiers have been probed and the claimant has been neither accepted nor rejected, the session terminates with an inconclusive decision and the claimant is referred to further manual interrogation. The method computes various performance characteristics such as the probability that a session of a legitimate claimant terminates inconclusively. These characteristics are valuable in the design of an effective record of identifiers for each of the claimants.