Granted Patent B1

US 12,400,146 · App. 17/717,527 · Granted Aug 26, 2025

Inventors: Harshil Shah (Centreville, VA); Manoj Kumar Goyal (Fremont, CA); Jin Yu (Union City, CA)

Assignee: CIPIO Inc.

G06N20/00

Abstract

Claims (46)

1. A method, comprising:

identifying a first set of data columns in a first data domain having a plurality of data types;

identifying a second set of data columns in a second data domain having a plurality of data types that are the same as or substantially similar to the plurality of data types of the first data domain;

selecting (1) a first subset of data columns from the first set of data columns in the first data domain and having a plurality of parameters, and (2) a second subset of data columns from the second set of data columns in the second data domain and having a plurality of parameters matching the plurality of parameters of the first subset of data columns;

arranging the first subset of data columns and the second set of data columns into a multi-domain dataset, each column in the multi-domain dataset including an attribute from the first set of data columns or an attribute the second set of data columns, each row in the multi-domain dataset including multi-domain data from the first data domain and the second data domain;

classifying each column in the multi-domain dataset into structured data or unstructured data, columns in the multi-domain dataset with structured data being mapped to a set of features for training at least one machine learning model, each feature from the set of features associated with the first data domain or the second data domain;

replacing a first set of under-performing features from a subset of features from the set of features in the second data domain, the under-performing features from the first set of under-performing features identified by evaluating the at least one machine learning model;

replacing a second set of under-performing features from a subset of features from the set of features in the first data domain, after the replacing the first set of under-performing features, the under-performing features from the second set of under-performing features identified by evaluating the at least one machine learning model, a predetermined portion of features from the subset of features from the set of features in the first data domain being retained; and

re-training the at least one machine learning model after replacing the first set of under-performing features and the second set of under-performing features.

2. The method of claim 1 , wherein the first data domain includes telecom data and the second data domain includes fitness data.

3. The method of claim 1 , wherein the multi-domain dataset includes a multi-domain profile.

4. The method of claim 1 , wherein the multi-domain dataset includes a multi-domain profile generated by concatenating rows in the multi-domain dataset having a common attribute.

5. The method of claim 1 , wherein the plurality of parameters of the first subset of data columns includes at least one of an email address or a phone number.

6. The method of claim 1 , wherein the first data domain includes at least one of a diet data domain, a medical record domain, a financial domain, a network domain, a demographic domain, or a customer domain.

7. The method of claim 1 , wherein the under-performing features from the first set of under-performing features decrease an accuracy of an output generated by the at least one machine learning model.

8. An apparatus, comprising:

a memory; and

a processor operatively coupled to the memory, the processor configured to:

identify a first set of data columns in a first data domain having a plurality of data types;

identify a second set of data columns in a second data domain having a plurality of data types that are the same as or substantially similar to the plurality of data types of the first data domain;

select (1) a first subset of data columns from the first set of data columns in the first data domain and having a plurality of parameters, and (2) a second subset of data columns from the second set of data columns in the second data domain and having a plurality of parameters matching the plurality of parameters of the first subset of data columns;

arrange the first subset of data columns and the second set of data columns into a multi-domain dataset, each column in the multi-domain dataset including an attribute from the first set of data columns or an attribute the second set of data columns, each row in the multi-domain dataset including multi-domain data from the first data domain and the second data domain;

classify each column in the multi-domain dataset into structured data or unstructured data, columns in the multi-domain dataset with structured data being mapped to a set of features for training at least one machine learning model, each feature from the set of features associated with the first data domain or the second data domain;

replace a first set of under-performing features from a subset of features from the set of features in the second data domain, the under-performing features from the first set of under-performing features identified by evaluating the at least one machine learning model;

replace a second set of under-performing features from a subset of features from the set of features in the first data domain, after the replacing the first set of under-performing features, the under-performing features from the second set of under-performing features identified by evaluating the at least one machine learning model, a predetermined portion of features from the subset of features from the set of features in the first data domain being retained; and

re-train the at least one machine learning model after replacing the first set of under-performing features and the second set of under-performing features.

9. The apparatus of claim 8 , wherein evaluating the under-performing features from the first set of under-performing features includes determining a speed the at least one machine learning model generating an output, the under-performing features from the first set of under-performing features identified based on the speed.

10. The apparatus of claim 8 , wherein the first data domain includes telecom data and the second data domain includes fitness data.

11. The apparatus of claim 8 , wherein the multi-domain dataset includes multi-domain profiles.

12. The apparatus of claim 8 , wherein the multi-domain dataset includes multi-domain profiles generated by concatenating rows in the multi-domain dataset having a common attribute.

13. The apparatus of claim 8 , wherein the plurality of parameters of the first subset of data columns includes at least one of an email address or a phone number.

14. The apparatus of claim 8 , wherein the first data domain includes at least one of a diet data domain, a medical record domain, a financial domain, a network domain, a demographic domain, or a customer domain.

15. The apparatus of claim 8 , wherein evaluating the under-performing features from the first set of under-performing features includes determining an accuracy of an output generated by the at least one machine learning model, the under-performing features from the first set of under-performing features identified based on the accuracy.

16. A non-transitory, processor readable medium storing instructions that, when executed by a processor, cause the processor to:

identify a first set of data columns in a first data domain having a plurality of data types;

identify a second set of data columns in a second data domain having a plurality of data types that are the same as or substantially similar to the plurality of data types of the first data domain;

select (1) a first subset of data columns from the first set of data columns in the first data domain and having a plurality of parameters, and (2) a second subset of data columns from the second set of data columns in the second data domain and having a plurality of parameters matching the plurality of parameters of the first subset of data columns;

arrange the first subset of data columns and the second set of data columns into a multi-domain dataset, each column in the multi-domain dataset including an attribute from the first set of data columns or an attribute the second set of data columns, each row in the multi-domain dataset including multi-domain data from the first data domain and the second data domain;

classify each column in the multi-domain dataset into structured data or unstructured data, columns in the multi-domain dataset with structured data being mapped to a set of features for training at least one machine learning model, each feature from the set of features associated with the first data domain or the second data domain;

replace a first set of under-performing features from a subset of features from the set of features in the second data domain, the under-performing features from the first set of under-performing features identified by evaluating the at least one machine learning model;

replace a second set of under-performing features from a subset of features from the set of features in the first data domain, after the replacing the first set of under-performing features, the under-performing features from the second set of under-performing features identified by evaluating the at least one machine learning model, a predetermined portion of features from the subset of features from the set of features in the first data domain being retained; and

re-train the at least one machine learning model after replacing the first set of under-performing features and the second set of under-performing features.

17. The non-transitory, processor readable medium of claim 16 , wherein the first data domain includes telecom data and the second data domain includes fitness data.

18. The non-transitory, processor readable medium of claim 16 , wherein the under-performing features from the first set of under-performing features decrease an accuracy of an output generated by the at least one machine learning model.

19. The non-transitory, processor readable medium of claim 16 , wherein the plurality of parameters of the first subset of data columns includes an email address and a phone number.

20. The non-transitory, processor readable medium of claim 16 , wherein the multi-domain dataset includes multi-domain profiles.

Assignments (2)

Continuity (1)

References Cited (20)