System and methods for privacy preserving cross-site federated learning

Patent metadata

Field	Value
Publication number	US-12045695-B2
Application number	US-202016804907-A
Country	US
Kind code	B2
Filing date	Feb 28, 2020
Priority date	Feb 28, 2020
Publication date	Jul 23, 2024
Grant date	Jul 23, 2024

Abstract

Data samples are transmitted from a central server to at least one local server apparatus. The central server receives a set of predictions from the at least one local server apparatus that are based on the transmitted set of data samples. The central server trains a central model based on the received set of predictions. The central model, or a portion of the central model corresponding to a task of interest, can then be sent to the at least one local server apparatus. Neither local data from local sites nor trained models from the local sites are transmitted to the central server. This ensures protection and security of data at the local sites.

First claim

What is claimed is:

1. A central server comprising:
a processor configured to:
transmit a set of public data samples from the central server with a central machine learning model to at least one local server apparatus, the at least one local server apparatus including a local machine learning model that is previously trained using private data samples, wherein the at least one local server apparatus is configured to generate a set of predictions based only on the set of public data samples;
cause the central machine learning model to receive only the set of predictions from the at least one local server apparatus; and
train the central machine learning model based on the received set of predictions that are based only on the public data samples, wherein the private data samples used to train the at least one local server apparatus are not accessible to the central machine learning model.

2. The central server according to claim 1 , wherein the processor is further configured to transmit at least a portion of the central model to the at least one local server apparatus.

3. The central server according to claim 2 , wherein the processor is further configured to determine at least one task of interest at the at least one local server apparatus, identify a portion of the central model corresponding to at least one task of interest, and transmit the identified portion of the central model corresponding to the at least one task of interest to the at least one local server apparatus.

4. The central server according to claim 1 , wherein the set of data samples is specific to a task of interest at the at least one local server apparatus.

5. The central server according to claim 1 , wherein the apparatus comprises a training/inference server.

6. The central server according to claim 1 , wherein the transmitted set of data samples corresponds to a task of interest at the at least one local server apparatus.

7. The central server according to claim 1 , wherein the processor is further configured to form an ensemble dataset corresponding to at least one task of interest from the received set of predictions.

8. A method, comprising:
transmitting from a central server, a set of public data samples to at least one local server apparatus, the at least one local server apparatus including a local machine learning model previously trained using private data samples;
receiving in the central server, a set of predictions from the at least one local server apparatus, the set of predictions generated by the local machine learning model using only the transmitted set of public data samples; and
training a central machine learning model in the central server based on the received set of predictions, wherein the private data samples used to train the at least one local server apparatus, or any products of the private data samples, are not transmitted to the central machine learning model.

9. The method according to claim 8 wherein the method further comprises transmitting at least a portion of the central model from the central server to the at least one local server apparatus.

10. The method according to claim 9 , the method further comprising determining at least one task of interest at the at least one local server apparatus, identifying a portion of the central model corresponding to at least one task of interest, and transmit the identified portion of the central model corresponding to the at least one task of interest to the at least one local server apparatus.

11. The method according to claim 8 , wherein the set of public data samples is specific to a task of interest at the at least one local server apparatus.

12. The method according to claim 8 , wherein the central server comprises a training/inference server.

13. The method according to claim 8 , wherein the central server is configured to aggregate products of inferences of individual ones of the at least one local server apparatus and train the central machine learning model using the aggregated products without using any local data from the at least one local server apparatus.

14. The method according to claim 8 , wherein the method further comprises forming an ensemble dataset corresponding to the task of interest from the received set of predictions.

15. The method according to claim 8 , the at least one local server apparatus is configured to generate the set of predictions using the local machine learning model which takes the public data samples as an input.

16. A computer program product comprising a non-transitory computer readable media having stored thereon program instructions that when executed by a processor causes the processor to perform the method according to claim 8 .

17. The central server according to claim 1 , wherein the set of predictions are generated by models on the at least one local server apparatus with data transmitted from the central machine learning model.

18. The central server according to claim 17 , wherein once the central server updates the central machine learning model, the update of the central machine learning model is transmitted back to the at least one local server apparatus, wherein the local machine learning model updates the set of predictions based on the update from the central machine learning model.

19. The central server according to claim 1 , wherein the at least one local server apparatus are trained independently of one another and the received set of predictions is based on the public data samples.

20. The method according to claim 8 , wherein the at least one local server apparatus are trained independently of one another and the received set of predictions is based on the public data samples.

Assignees

• Shanghai United Imaging Intelligence Co Ltd

Inventors

• Karanam Srikrishna
• Wu Ziyan
• Sharma Abhishek
• Innanje Arun
• Chen Terrence

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06N3/098

  Distributed learning, e.g. federated learning · CPC title

• H04L63/20

  for managing network security; network security policies in general (filtering policies H04L63/0227) · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/088

  Non-supervised learning, e.g. competitive learning · CPC title