Machine-learning system for optimising the performance of a biometric system

A computer-implemented method for adaptively changing a matching threshold in a biometric system to verify an identity of a subject using machine-learning, wherein the matching threshold is adapted to changing environmental conditions, the method comprising:

• receiving, by a network device, input data from the biometric system via a communication interface,
• the communication interface being configured to communicatively couple the network device with the biometric system and an environmental sensor;
• the input data from the biometric system comprising biometric identification data samples received from a biometric sensor associated with the biometric system,
• the biometric sensor being associated with at least one of:
• a fingerprint reader,
• a facial recognition system,
• a retinal scanner, or
• an iris scanner,
• receiving, by the network device and via the communication interface, input data from the environmental sensor,
• the input data from the environmental sensor comprising an environmental measure determined by the environmental sensor;
• aggregating, by the network device, a batch of the received input data from the biometric system and the environmental sensor;
• computing, by the network device, a first suggestion for the matching threshold of the biometric system based on the aggregated batch and based on pre-determined identity information of the subject,
• the first suggestion for the matching threshold being computed by using a first internal model, and
• the first internal model comprising a first machine-learning algorithm executed by the network device;
• computing, by the network device, a second suggestion for the matching threshold of the biometric system based on the aggregated batch and based on the pre-determined identity information of the subject,
• second suggestions being computed by using a second internal model,
• the second internal model comprising a second machine-learning algorithm executed by the network device and different from the first machine-learning algorithm used by the first internal model;
• comparing, by the network device, the first suggestion for the matching threshold with the second suggestion for the matching threshold to generate a final suggestion for the matching threshold of the biometric system based on the comparison;
• outputting, by the network device, the final suggestion for the matching threshold of the biometric system via the communication interface;
• receiving, by the network device, a matching score, associated with the identity of the subject, from the biometric system based on the final suggestion for the matching threshold and the pre-determined identity information of the subject;
• authenticating, by the network device, the subject when the matching score indicates an identity match;
• and
• refusing, by the network device, to authenticate the subject when the matching score indicates no identity match.

A non-transitory computer-readable medium storing instructions, the instructions comprising:

• one or more instructions that, when executed by a network device, cause the network device to:
• receive input data from a biometric system via a communication interface,
• the communication interface configured to communicatively couple the machine-learning system with the biometric system and an environmental sensor,
• wherein the biometric system is to verify an identity of a subject,
• wherein the input data from the biometric system comprises biometric identification data samples received from a biometric sensor associated with the biometric system,
• wherein the biometric sensor is associated with at least one of:
• a fingerprint reader,
• a facial recognition system,
• a retinal scanner, or
• an iris scanner,
• receive, via the communication interface, input data from the environmental sensor,
• the input data from the environmental sensor comprising an environmental measure determined by the environmental sensor;
• aggregate a batch of the received input data from the biometric system and from the environmental sensor;
• compute a first suggestion for a matching threshold of the biometric system based on the aggregated batch and based on pre-determined identity information of the subject,
• wherein computing the first suggestion for the matching threshold is performed using a first internal model, and
• wherein the first internal model comprises a first machine-learning algorithm executed by the network device;
• compute a second suggestion for the matching threshold of the biometric system based on the aggregated batch and based on the pre-determined identity information of the subject,
• wherein computing the second suggestion uses a second internal model,
• wherein the second internal model comprises a second machine-learning algorithm executed by the network device and different from the first machine-learning algorithm used by the first internal model;
• compare the first suggestion for the matching threshold with the second suggestion for the matching threshold to generate a final suggestion for the matching threshold of the biometric system based on the comparison;
• output the final suggestion for the matching threshold of the biometric system via the communication interface;
• receive a matching score, associated with the identity of the subject, from the biometric system based on the final suggestion for the matching threshold and the pre-determined identity information of the subject;
• authenticate the subject when the matching score indicates an identity match;
• and
• refuse to authenticate the subject when the matching score indicates no identity match.