Device-side federated machine learning computer system architecture

Patent metadata

Field	Value
Publication number	US-12081541-B2
Application number	US-202117395204-A
Country	US
Kind code	B2
Filing date	Aug 5, 2021
Priority date	Aug 5, 2021
Publication date	Sep 3, 2024
Grant date	Sep 3, 2024

Abstract

Techniques are disclosed in which a computer system receives a transaction request and uses a federated machine learning model to analyze the transaction request. A server computer system may generate a federated machine learning model and distribute portions of the federated machine learning models to other components of the computer system including a user device and/or edge servers. In various embodiments, various components of the computer system apply transaction request evaluation factors to the portions of the federated machine learning model to generate scores. The server computer system uses the scores to determine a response to the transaction request.

First claim

What is claimed is:

1. A user device comprising:
a computer processor circuit; and
a computer memory circuit storing instructions that when executed by the computer processor circuit cause the user device to perform operations including:
receiving a user device portion of a federated machine learning model that was generated by a server computer system using a dataset of previous transaction requests, wherein the federated machine learning model includes the user device portion and a server portion;
receiving a transaction request from a user;
collecting (a) user behavior information about how the user has used the user device and (b) transaction information about a transaction requested by the transaction request;
generating, using the user device portion of the federated machine learning model and the user behavior information, one or more user device scores by applying the user behavior information to the federated machine learning model;
sending, to the server computer system, an indication of the transaction request, the one or more user device scores, and the transaction information; and
receiving, from the server computer system, a response to the transaction request.

2. The user device of claim 1 , wherein the operations further include:
training the user device portion using the user behavior information; and
wherein generating the one or more user device scores includes generating an indication of whether user behavior in a period of time before receiving the transaction request has deviated from a previous trend in user behavior.

3. The user device of claim 1 ,
wherein the collecting includes collecting (c) user device information;
wherein generating the one or more user device scores includes applying the user device information to the federated machine learning model; and
wherein the user behavior information and device information used to generate the one or more user device scores is not sent from the user device to another computer system.

4. The user device of claim 3 , wherein the operations further include:
sending, from the user device to an edge server, user behavior information and device information that was not used to generate the one or more user device scores.

5. The user device of claim 1 ,
wherein the one or more user device scores are generated periodically and independent from receiving the transaction request.

6. The user device of claim 1 ,
wherein the collecting includes (c) collecting biometric information about the user of the user device;
wherein generating the one or more user device scores includes applying indications of the biometric information to the federated machine learning model; and
wherein biometric information is not sent from the user device to another computer system.

7. The user device of claim 1 , wherein the operations further comprise:
receiving a step-up challenge from the server computer system;
receiving a solution to the step-up challenge from the user; and
sending an indication of the solution to the step-up challenge to the server computer system.

8. A method comprising:
storing, at a remote computer system, a remote portion of a federated machine learning model that was generated by a server computer system using a dataset of previous transaction requests, wherein the federated machine learning model includes the remote portion and a server portion;
receiving, by the remote computer system, a transaction request from a user;
receiving, by the remote computer system, transaction request evaluation factors, wherein the transaction request evaluation factors include information that is usable to identify the user;
generating, by the remote computer system using the remote portion of the federated machine learning model and a first subset of the transaction request evaluation factors, one or more remote computer system scores;
sending, from the remote computer system to the server computer system, an indication of the transaction request, the one or more remote computer system scores, and a second subset of the transaction request evaluation factors; and
receiving, from the server computer system, a response to the transaction request.

9. The method of claim 8 ,
wherein the remote computer system is a user device; and
wherein receiving the transaction request evaluation factors includes using the user device to collect the transaction request evaluation factors.

10. The method of claim 9 ,
wherein the user device includes a safe list of software approved to collect and analyze information that is usable to identify a user without permission from the user;
wherein the transaction request evaluation factors are collected by authentication software running on the user device;
wherein the first subset of the transaction request evaluation factors includes information that is usable to identify the user; and
wherein at least some of the information that is usable to identify the user is not sent from the user device to another computer system.

11. The method of claim 10 , further comprising:
sending, from the user device to an edge server, a portion of the information that is usable to identify the user.

12. The method of claim 8 ,
wherein the remote computer system is an edge server; and
wherein receiving the transaction request evaluation factors includes receiving the transaction request evaluation factors from a user device.

13. The method of claim 12 ,
wherein the federated machine learning model includes a user device portion that was previously sent to the user device via the edge server;
the method further comprising:
receiving, at the edge server from the user device, one or more user device scores generated by the user device using the user device portion of the federated machine learning model and a third subset of transaction request evaluation factors that was collected by the user device but not sent to the edge server; and
sending, from the edge server to the server computer system, the one or more user device scores.

14. The method of claim 12 ,
wherein the first subset of the transaction request evaluation factors is not sent from the edge server to the server computer system.

15. The method of claim 8 , wherein the response to the transaction request is sending a step-up challenge, the method further comprising:
receiving, at the remote computer system, a step-up challenge from the server computer system;
receiving, at the remote computer system a solution to the step-up challenge from the user; and
sending an indication of the solution to the step-up challenge to the server computer system.

16. A method comprising:
storing, at a user device, a user device portion of a federated machine learning model that was generated by a server computer system using a dataset of previous transaction requests, wherein the federated machine learning model includes the user device portion and a server portion;
receiving, at the user device, a transaction request from a user;
collecting, by the user device, (a) user behavior information about how the user has used the user device, (b) user device information, and (c) transaction information about a transaction requested by the transaction request;
generating, by the user device using the user device portion of the federated machine learning model and the user behavior information, one or more user device scores by applying the user behavior information and user device information to the federated machine learning model;
sending, from the user device to the server computer system, an indication of the transaction

Assignees

• Paypal Inc

Inventors

• Chhibber Abhishek
• Desai Darshankumar Bhadrasinh
• Todasco Michael Charles
• Naware Vidyut Mukund
• Sharma Nitin S

Classifications

• H04L63/0876

  based on the identity of the terminal or configuration, e.g. MAC address, hardware or software configuration or device fingerprint · CPC title

• G06N20/20

  Ensemble learning · CPC title

• G06N20/00

  Machine learning · CPC title

• H04L2463/082

  applying multi-factor authentication · CPC title

• H04L63/0861 Primary

  using biometrical features, e.g. fingerprint, retina-scan (cryptographic mechanisms or cryptographic arrangements for entity authentication using biological data H04L9/3231) · CPC title