Managing deviations between expected and normal operations of authentication systems
There are disclosed techniques for use in authentication. In one embodiment, the techniques comprise generating first and second distributions. The first distribution relating to risk scores expected to be produced by an authentication system in connection with requests to access a computerized resource. The expected risk scores are based on a normalization process configured to produce risk scores by normalizing raw risk scores in connection with requests. The second distribution relates to risk scores actually produced by the authentication system in connection with requests. The actual risk scores include risk scores normalized by the normalization process. The techniques also comprise comparing the first and second distributions by determining a Kolmogorov-Smirnov distance between the respective distributions. The techniques also comprise initiating, based on the comparison, a failover of the normalization process to a new normalization process for use by the authentication system.
1. A computer-implemented method, comprising:
generating, by processing circuitry, a first distribution of risk scores expected to be produced by an authentication system in connection with a first set of one or more requests to access a computerized resource, wherein generating the first distribution comprises utilizing a normalization process to produce the expected risk scores by normalizing preliminary expected risk scores in connection with the first set of one or more requests to a scale of risk scores such that a pre-defined amount of the first set of one or more requests have expected risk scores in a range of risk scores at an end of the scale indicating high risk;
generating, by processing circuitry, a second distribution relating to risk scores actually produced by the authentication system in connection with a second set of one or more requests to access the computerized resource, wherein generating the second distribution comprises utilizing a normalization process to produce the actual risk scores by normalizing preliminary actual risk scores in connection with the second set of one or more requests to the scale of risk scores;
performing, by processing circuitry, a comparison between the first and the second distributions, wherein performing the comparison comprises utilizing the Kolmogorov-Smirnov distance to detect a deviation between the expected and the actual risk scores in the range of risk scores at the end of the scale indicating high risk;
performing, by processing circuitry, a normalization process failover to a new normalization process after detection of the deviation, wherein the new normalization process is configured to reduce the deviation between the expected and the actual risk scores in the range of risk scores at the end of the scale indicating high risk, wherein the new normalization process is based on data related to a third set of one or more requests to access the computerized resource that are received after the detection of the deviation;
receiving, by processing circuitry, a fourth set of one or more requests to access the computerized resource after the detection of the deviation,
determining, by processing circuitry, actual risk scores in connection with the fourth set of one or more requests by normalizing preliminary actual risk scores in connection with the fourth set of one or more requests to the scale of risk scores in accordance with the new normalization process; and
utilizing, by processing circuitry, the actual risk scores in connection with the fourth set of one or more requests to control access to the computerized resource.
2. The method as claimed in claim 1 , wherein the first distribution comprises a first empirical cumulative distribution suitable for comparing with a distribution that relates to actual risk scores.
3. The method as claimed in claim 1 , wherein the second distribution comprises a second empirical cumulative distribution suitable for comparing with a distribution that relates to expected risk scores.
4. The method as claimed in claim 1 , wherein the initiation of the failover is based on the Kolmogorov-Smirnov distance between the respective distributions being larger than a value that is based on Kolmogorov-Smirnov distances associated with historical distributions.
5. The method as claimed in claim 1 , wherein the initiation of the failover is based on the Kolmogorov-Smirnov distance between the respective distributions being larger than a conversion function associated with a required assurance level of a goodness-of-fit test.
6. The method as claimed in claim 1 , wherein the initiation of the failover is based on there being a sufficient amount of historical information.
7. An apparatus, comprising:
memory; and
processing circuitry coupled to the memory, the memory storing instructions which, when executed by the processing circuitry, cause the processing circuitry to:
generate a first distribution of risk scores expected to be produced by an authentication system in connection with a first set of one or more requests to access a computerized resource, wherein generating the first distribution comprises utilizing a normalization process to produce the expected risk scores by normalizing preliminary expected risk scores in connection with the first set of one or more requests to a scale of risk scores such that a pre-defined amount of the first set of one or more requests have expected risk scores in a range of risk scores at an end of the scale indicating high risk;
generate a second distribution relating to risk scores actually produced by the authentication system in connection with a second set of one or more requests to access the computerized resource, wherein generating the second distribution comprises utilizing a normalization process to produce the actual risk scores by normalizing preliminary actual risk scores in connection with the second set of one or more requests to the scale of risk scores;
perform a comparison between the first and the second distributions, wherein performing the comparison comprises utilizing the Kolmogorov-Smirnov distance to detect a deviation between the expected and the actual risk scores in the range of risk scores at the end of the scale indicating high risk;
perform a normalization process failover to a new normalization process after detection of the deviation, wherein the new normalization process is configured to reduce the deviation between the expected and the actual risk scores in the range of risk scores at the end of the scale indicating high risk, wherein the new normalization process is based on data related to a third set of one or more requests to access the computerized resource that are received after the detection of the deviation;
receive a fourth set of one or more requests to access the computerized resource after the detection of the deviation;
determine actual risk scores in connection with the fourth set of one or more requests by normalizing preliminary actual risk scores in connection with the fourth set of one or more requests to the scale of risk scores in accordance with the new normalization process; and
utilize actual risk scores in connection with the fourth set of one or more requests to control access to the computerized resource.
8. The apparatus as claimed in claim 7 , wherein the first distribution comprises a first empirical cumulative distribution suitable for comparing with a distribution that relates to actual risk scores.
9. The apparatus as claimed in claim 7 , wherein the second distribution comprises a second empirical cumulative distribution suitable for comparing with a distribution that relates to expected risk scores.
10. The apparatus as claimed in claim 7 , wherein the initiation of the failover is based on the Kolmogorov-Smirnov distance between the respective distributions being larger than a value that is based on Kolmogorov-Smirnov distances associated with historical distributions.
11. The apparatus as claimed in claim 7 , wherein the initiation of the failover is based on the Kolmogorov-Smirnov distance between the respective distributions being larger than a conversion function associated with a required assurance level of a goodness-of-fit test.
12. The apparatus as claimed in claim 7 , wherein the initiation of the failover is based on there being a sufficient amount of historical information.
13. A computer program product having a non-transitory computer-readable medium storing instructions, the instructions, when carried out by processing circuitry, causing the processing circuitry to perform a method of:
generating a first distribution of risk scores expected to be produced by an authentication system in connection with a first set of one or more requests to access a computerized resource, wherein generating the first distribution comprises utilizing a normalization process to produce the expected risk scores by normalizing preliminary expected risk scores in connection with the first set of one or more requests to a scale of risk scores such that a pre-defined amount of the first set of one or more requests have expected risk scores in a range of risk scores at an end of the scale indicating high risk;
generating a second distribution relating to risk scores actually produced by the authentication system in connection with a second set of one or more requests to access the computerized resource, wherein generating the second distribution comprises utilizing a normalization process to produce the actual risk scores by normalizing preliminary actual risk scores in connection with the second set of one or more requests to the scale of risk scores;
performing a comparison between the first and the second distributions, wherein performing the comparison comprises utilizing the Kolmogorov-Smirnov distance to detect a deviation between the expected and the actual risk scores in the range of risk scores at the end of the scale indicating high risk;
performing a normalization process failover to a new normalization process after detection of the deviation, wherein the new normalization process is configured to reduce the deviation between the expected and the actual risk scores in the range of risk scores at the end of the scale indicating high risk, wherein the new normalization process is based on data related to a third set of one or more requests to access the computerized resource that are received after the detection of the deviation;
receiving a fourth set of one or more requests to access the computerized resource after the detection of the deviation;
determining actual risk scores in connection with the fourth set of one or more requests by normalizing preliminary actual risk scores in connection with the fourth set of one or more requests to the scale of risk scores in accordance with the new normalization process; and
utilizing the actual risk scores in connection with the fourth set of one or more requests to control access to the computerized resource.
14. The computer program product as claimed in claim 13 , wherein the first distribution comprises a first empirical cumulative distribution suitable for comparing with a distribution that relates to actual risk scores.
15. The computer program product as claimed in claim 13 , wherein the second distribution comprises a second empirical cumulative distribution suitable for comparing with a distribution that relates to expected risk scores.
16. The computer program product as claimed in claim 13 , wherein the initiation of the failover is based on the Kolmogorov-Smirnov distance between the respective distributions being larger than a value that is based on Kolmogorov-Smirnov distances associated with historical distributions.
17. The computer program product as claimed in claim 13 , wherein the initiation of the failover is based on the Kolmogorov-Smirnov distance between the respective distributions being larger than a conversion function associated with a required assurance level of a goodness-of-fit test.
18. The computer program product as claimed in claim 13 , wherein the initiation of the failover is based on there being a sufficient amount of historical information.