US 2024/0193431 A1Published application

Methods and systems for forkable federated learning pathways for versatile learning pathways

AssigneeCapital One Services, LLC

InventorsMichael DAVIS, Taylor TURNER, Tyler FARNAN, Kenny BEAN, Jeremy GOODSITT

Life of this patent

Estimating the term…

Loading patent…

Abstract

Methods and systems are described for novel uses and/or improvements to federated learning. As one example, methods and systems are described for improving the applicability of federated learning across various applications and increasing the efficiency of training a global model through federated learning. As another example, methods and systems are described for ensuring comprehensive training data is available to models assigned by the federated learning server. Additionally, methods and systems are described for improving the rate of training a global model through federated learning.

Claims20

Dependent claims

A system for forking federated learning pathways for improved versatility, the system comprising:

one or more processors;
and
a non-transitory computer readable medium having instructions recorded thereon that when executed by the one or more processors cause operations comprising:
receiving, by a federated learning model being executed on a remote server, a first aggregated data stream, wherein the first aggregated data stream receives data from a plurality of user devices, wherein the federated learning model is trained on the first aggregated data stream;
selecting a first subset of the plurality of user devices, wherein the federated learning model sends the first subset of the plurality of user devices a shared model to train;
generating a first fork in the first aggregated data stream by generating a second aggregated data stream at a first time, wherein the second aggregated data stream receives data from the first subset of the plurality of user devices;
generating a first fork identifier based on the first time and the first subset of the plurality of user devices;
training a first machine learning model on the second aggregated data stream, wherein training is conducted locally on the plurality of user devices;
receiving a first user request to merge the first machine learning model into the federated learning model;
in response to the first user request, determining a first weight for merging the first machine learning model into the federated learning model based on the first fork identifier;
and
merging the first machine learning model into the federated learning model based on the first weight wherein merging the first machine learning model into the federated learning model includes merging effective parameters from each model, wherein effective parameters are determined by using one or more optimization algorithms.

9 more elements

A method for forking federated learning pathways for improved versatility, the method comprising:

receiving, by a federated learning model being executed on a remote server, a first aggregated data stream, wherein the first aggregated data stream receives data from a plurality of user devices, wherein the federated learning model is trained on the first aggregated data stream;
selecting a first subset of the plurality of user devices;
generating a first fork in the first aggregated data stream by generating a second aggregated data stream at a first time, wherein the second aggregated data stream receives data from the first subset of the plurality of user devices;
generating a first fork identifier based on the first time and the first subset of the plurality of user devices;
training a first machine learning model on the second aggregated data stream;
receiving a first user request to merge the first machine learning model into the federated learning model;
in response to the first user request, determining a first weight for merging the first machine learning model into the federated learning model based on the first fork identifier;
and
merging the first machine learning model into the federated learning model based on the first weight.

6 more elements

A non-transitory, computer-readable medium comprising instructions recorded thereon that when executed by one or more processors causes operations comprising:

receiving, by a federated learning model being executed on a remote server, a first aggregated data stream, wherein the first aggregated data stream receives data from a plurality of user devices, wherein the federated learning model is trained on the first aggregated data stream;
selecting a first subset of the plurality of user devices;
generating a first fork in the first aggregated data stream by generating a second aggregated data stream at a first time, wherein the second aggregated data stream receives data from the first subset of the plurality of user devices;
generating a first fork identifier based on the first time and the first subset of the plurality of user devices;
training a first machine learning model on the second aggregated data stream;
receiving a first user request to merge the first machine learning model into the federated learning model;
in response to the first user request, determining a first weight for merging the first machine learning model into the federated learning model based on the first fork identifier;
and
merging the first machine learning model into the federated learning model based on the first weight.

6 more elements

Dependencies are inferred from the claim text.