Granted Patent B2

US 12,530,623 · App. 18/094,679 · Granted Jan 20, 2026

Inventors: Danial Dervovic (London, GB); Freddy Lecue (Mamaroneck, NY); Carlos Perez (London, GB); Pietro Smacchia (London, GB); Daniele Magazzeni (London, GB)

Assignee: JPMORGAN CHASE BANK, N.A.

G06N20/00

Abstract

Claims (66)

1 . A method for computing change-agnostic data points by utilizing one or more processors along with allocated memory, the method comprising:

receiving at least a first set of raw data that is usable for training a machine learning model;

training the machine learning model by using the at least the first set of raw data;

computing a set of explanations for all combinations based on output data of the trained machine learning model, the first set of raw data, and sampled raw data computed by applying a sampling algorithm on the raw data;

computing a compact representation of the set of explanations corresponding to a pre-configured dimension based on compression quality and generating a set of compressed explanations;

computing a unique representation of model explanation with respect to the pre-configured dimension;

determining whether the model explanation is robust to changes in data through data perturbation; and

computing change-agnostic data points based on determining that the model explanation is robust to changes in data through data perturbation.

2 . The method of claim 1 , wherein the first set of raw data includes prediction data computed by the machine learning model, input raw data utilized for navigating explanation-dimension used for extending explanation search space, and sampled raw data computed by implementing a data distribution sampling strategy on the input raw data.

3 . The method according to claim 1 , wherein the preconfigured dimension includes one dimensional representation, two dimensional representation, three dimensional representation, or more than three dimensional representations based on the compression quality.

4 . The method according to claim 3 , wherein in computing the set of compressed explanations, the method further comprising:

implementing an auto-encoder algorithm or principal component analysis algorithm.

5 . The method according to claim 1 , wherein in determining whether the model explanation is robust to changes in data through data perturbation, the method further comprising:

receiving perturbated input data to be tested, wherein the perturbated input data is generated by implementing a data perturbation strategy;

receiving the first set of raw data;

receiving the sampled raw data;

generating perturbated target data based on processing the received perturbated input data, the first set of raw data, and the sampled raw data; and

performing a variability test based on compressed explanations of the perturbated target data, compressed explanations of input target data; and explanation variability score that is described as a vector along the pre-configured dimension.

6 . The method according to claim 5 , wherein the data perturbation strategy includes adding 1 to all features or replace one feature by an average value of all individual features to the at least the first set of raw data.

7 . The method according to claim 6 , further comprising:

determining whether a numerical value of compressed explanations of the perturbated target data is greater than the explanation variability score; and

computing the change-agnostic data points based on a determination that the numerical value of the compressed explanations of the perturbated target data is greater than the explanation variability score.

8 . The method according to claim 1 , wherein the machine learning model includes one or more of the following models: decision tree, ensemble trees, neural network architectures, and predictive model.

9 . A system for computing change-agnostic data points, the system comprising:

a processor; and

a memory operatively connected to the processor via a communication interface, the memory storing computer readable instructions, when executed, causes the processor to:

receive at least a first set of raw data that is usable for training a machine learning model;

train the machine learning model by using the at least the first set of raw data;

compute a set of explanations for all combinations based on output data of the trained machine learning model, the first set of raw data, and sampled raw data computed by applying a sampling algorithm on the raw data;

compute a compact representation of the set of explanations corresponding to a pre-configured dimension based on compression quality and generating a set of compressed explanations;

compute a unique representation of model explanation with respect to the pre-configured dimension;

determine whether the model explanation is robust to changes in data through data perturbation; and

compute change-agnostic data points based on determining that the model explanation is robust to changes in data through data perturbation.

10 . The system of claim 9 , wherein the first set of raw data includes prediction data computed by the machine learning model, input raw data utilized for navigating explanation-dimension used for extending explanation search space, and sampled raw data computed by implementing a data distribution sampling strategy on the input raw data.

11 . The system according to claim 9 , wherein the preconfigured dimension includes one dimensional representation, two dimensional representation, three dimensional representation, or more than three dimensional representations based on the compression quality.

12 . The system according to claim 11 , in computing the set of compressed explanations, the processor is further configured to:

implement an auto-encoder algorithm or principal component analysis algorithm.

13 . The system according to claim 9 , in determining whether the model explanation is robust to changes in data through data perturbation, the processor is further configured to:

receive perturbated input data to be tested, wherein the perturbated input data is generated by implementing a data perturbation strategy;

receive the first set of raw data;

receive the sampled raw data;

generate perturbated target data based on processing the received perturbated input data, the first set of raw data, and the sampled raw data; and

perform a variability test based on compressed explanations of the perturbated target data, compressed explanations of input target data; and explanation variability score that is described as a vector along the pre-configured dimension.

14 . The system according to claim 13 , wherein the data perturbation strategy includes adding 1 to all features or replace one feature by an average value of all individual features to the at least the first set of raw data.

15 . The system according to claim 14 , wherein the processor is further configured to:

determine whether a numerical value of compressed explanations of the perturbated target data is greater than the explanation variability score; and

compute the change-agnostic data points based on a determination that the numerical value of the compressed explanations of the perturbated target data is greater than the explanation variability score.

16 . The system according to claim 9 , wherein the machine learning model includes one or more of the following models: decision tree, ensemble trees, neural network architectures, and predictive model.

17 . A non-transitory computer readable medium configured to store instructions for computing change-agnostic data points, the instructions, when executed, cause a processor to perform the following:

receiving at least a first set of raw data that is usable for training a machine learning model;

training the machine learning model by using the at least the first set of raw data;

computing a set of explanations for all combinations based on output data of the trained machine learning model, the first set of raw data, and sampled raw data computed by applying a sampling algorithm on the raw data;

computing a compact representation of the set of explanations corresponding to a pre-configured dimension based on compression quality and generating a set of compressed explanations;

computing a unique representation of model explanation with respect to the pre-configured dimension;

determining whether the model explanation is robust to changes in data through data perturbation; and

computing change-agnostic data points based on determining that the model explanation is robust to changes in data through data perturbation.

18 . The non-transitory computer readable medium according to claim 17 , in determining whether the model explanation is robust to changes in data through data perturbation, the instructions, when executed, cause the processor to further perform the following:

receiving perturbated input data to be tested, wherein the perturbated input data is generated by implementing a data perturbation strategy;

receiving the first set of raw data;

receiving the sampled raw data;

generating perturbated target data based on processing the received perturbated input data, the first set of raw data, and the sampled raw data; and

performing a variability test based on compressed explanations of the perturbated target data, compressed explanations of input target data; and explanation variability score that is described as a vector along the pre-configured dimension.

19 . The non-transitory computer readable medium according to claim 18 , wherein the data perturbation strategy includes adding 1 to all features or replace one feature by an average value of all individual features to the at least the first set of raw data.

20 . The non-transitory computer readable medium according to claim 19 , the instructions, when executed, cause the processor to further perform the following:

determining whether a numerical value of compressed explanations of the perturbated target data is greater than the explanation variability score; and

computing the change-agnostic data points based on a determination that the numerical value of the compressed explanations of the perturbated target data is greater than the explanation variability score.

Assignments (1)

Continuity (1)

References Cited (26)