IP Library Granted Patent US 12705389
Granted Patent B2
US 12705389 · App. 17/993,634 · Granted Aug 11, 2026

Mitigating private data leakage in a federated learning system

Inventors: Jerome Anand (Bangalore, IN); Anil Kumar Bangalore Rajashekar (Bengaluru, IN); Dinkar Dhawale (Bengaluru, IN); Sachin P. Kamat (Bangalore, IN); Shidlingeshwar Khatakalle (Gadhinglaj, IN); Mitul Shah (Bengaluru, IN)
Assignee: INTEL CORPORATION
G06F21/6254G06N20/00
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12705389
App. No.
17/993,634
Filed
Nov 23, 2022
Granted
Aug 11, 2026
Kind
B2
Art Unit
2431
USPC
726/27
Abstract

Protecting data privacy in a federated learning computing environment includes receiving a model trained by a federated server with public data using global model parameters, getting local shareable data from a local shareable database, training the model with the local shareable data using the global model parameters to generate local model parameters, and obscuring the local model parameters. Protecting data privacy includes sending the local model parameters to the federated server, modifying the model to classify private data, and training the model with the private data using the local model parameters and updating the local model parameters.

Claims (28)

1 . A method comprising:

receiving, by a computing device, a model trained with public data based on global model parameters, wherein the model is received from a federated server device;

training the model with private data using the local model parameters as model parameters and updating the local model parameters;

retraining the trained model with private data using updated local model parameters as initialization parameters;

obscuring the updated local model parameters to secure the private data to protect privacy of local shareable data, wherein the local shareable data is received from a local shareable database, wherein obscuring the local model parameters comprises adding noise data to the local model parameters to mitigate potential statistical inferences of the private data associated with the local model parameters; and

transmitting, to the federated server device, the retrained model based on the obscured updated local model parameters such that the private data remains classified.

2 . The method of claim 1 , wherein the retrained model to classify the private data, wherein the public data is stored in a public database accessible by the federated server device, and wherein the private data is stored in a private database accessible by the computing device having authorization to access the private data, wherein the model comprises a machine-learning model.

3 . The method of claim 1 , wherein modifying the retrained model to classify the private data comprises adding a new classifier layer to the retrained model to identify a pattern associated with the private data.

4 . The method of claim 1 , wherein the computing device comprises processing circuitry coupled to a memory, the processing circuitry having application processing circuitry or graphics processing circuitry.

5 . A computing device comprising:

processing circuitry coupled to a memory, the processing circuitry to:

receive a model trained with public data based on global model parameters, wherein the model is received from a federated server device;

train the model with private data using the local model parameters as model parameters and updating the local model parameters;

retrain the trained model with private data using local model parameters as initialization parameters;

obscure the updated local model parameters to secure the private data to protect privacy of local shareable data, wherein the local shareable data is received from a local shareable database, wherein obscuring the local model parameters comprises adding noise data to the local model parameters to mitigate potential statistical inferences of the private data associated with the local model parameters; and

transmit, to the federated server device, the retrained model based on the obscured updated local model parameters such that the private data remains classified.

6 . The computing device of claim 5 , wherein the retrained model to classify the private data, wherein the public data is stored in a public database accessible by the federated server device, and wherein the private data is stored in a private database accessible by the computing device having authorization to access the private data, wherein the model comprises a machine-learning model.

7 . The computing device of claim 5 , wherein modifying the retrained model to classify the private data comprises adding a new classifier layer to the retrained model to identify a pattern associated with the private data.

8 . The computing device of claim 5 , wherein the processing circuitry comprises application processing circuitry or graphics processing circuitry.

9 . At least one non-transitory computer-readable medium having stored thereon instructions which, when executed, cause a computing device to perform operations comprising:

receiving a model trained with public data based on global model parameters, wherein the model is received from a federated server device;

training the model with private data using the local model parameters as model parameters and updating the local model parameters;

retraining the trained model with private data using local model parameters as initialization parameters;

obscuring the updated local model parameters to secure the private data to protect privacy of local shareable data, wherein the local shareable data is received from a local shareable database, wherein obscuring the local model parameters comprises adding noise data to the local model parameters to mitigate potential statistical inferences of the private data associated with the local model parameters; and

transmitting, to the federated server device, the retrained model based on the obscured updated local model parameters such that the private data remains classified.

10 . The non-transitory computer-readable medium of claim 9 , wherein the retrained model to classify the private data, wherein the public data is stored in a public database accessible by the federated server device, and wherein the private data is stored in a private database accessible by the computing device having authorization to access the private data, wherein the model comprises a machine-learning model.

11 . The non-transitory computer-readable medium of claim 9 , wherein modifying the retrained model to classify the private data comprises adding a new classifier layer to the retrained model to identify a pattern associated with the private data.

12 . The non-transitory computer-readable medium of claim 9 , wherein the computing device comprises processing circuitry having application processing circuitry or graphics processing circuitry.