US 11,941,118 B2
System and method to build robust classifiers against evasion attacks
Devu Manikantan Shila, West Hartford, CT (US)
Assigned to CARRIER CORPORATION, Palm Beach Gardens, FL (US)
Appl. No. 16/972,243
Filed by CARRIER CORPORATION, Palm Beach Gardens, FL (US)
PCT Filed Oct. 30, 2019, PCT No. PCT/US2019/058697
§ 371(c)(1), (2) Date Dec. 4, 2020,
PCT Pub. No. WO2020/096826, PCT Pub. Date May 14, 2020.
Claims priority of application No. 201811041945 (IN), filed on Nov. 6, 2018.
Prior Publication US 2021/0256121 A1, Aug. 19, 2021
Int. Cl. G06F 21/56 (2013.01); G06F 18/2113 (2023.01); G06F 18/22 (2023.01); G06F 21/55 (2013.01); G06N 20/00 (2019.01)
CPC G06F 21/562 (2013.01) [G06F 18/2113 (2023.01); G06F 18/22 (2023.01); G06F 21/554 (2013.01); G06N 20/00 (2019.01)] 13 Claims
OG exemplary drawing
 
1. A system for building a robust classifier against evasion attacks, the system comprising:
a storage medium, the storage medium being coupled to a processor;
the processor configured to:
receive an application;
identify one or more features of the application;
determine a first confidence score for a first version of the application including a first set of features and determining a second confidence score for a second version of the application including a second set of features, wherein the first set of features is different than the second set of features;
determine a difference between the first confidence score and the second confidence score;
compare the difference with a convergence threshold;
based on the comparison, determine whether the first confidence score exceeds a confidence score threshold; and
generate a report based on determining the first confidence score exceeds the confidence score threshold;
wherein in response to determining the difference is greater than the convergence threshold, the processor is configured to:
determine a third version of the application by removing one feature of the one or more features from the second version of the application;
determine a third confidence score of the third version of the application;
determine a difference between the second confidence score and the third confidence score;
compare the difference with the convergence threshold;
based on the comparison, determine whether the second confidence score exceeds the confidence score threshold; and
generate the report based on determining the second confidence score exceeds the confidence score threshold.