US 12,614,103 B1Granted

Distributed training of machine learning models in wireless computing environments

AssigneeT-Mobile USA, Inc.

InventorGaviphat Lekutai

Life of this patent

Estimating the term…

Loading patent…

Abstract

An artificial intelligence or machine learning controller generates aggregate machine learning models using models trained by a set of wireless computing devices. A first computing device accesses a set of training data that comprises a set of input data points and a set of output data points. The first computing device trains a first machine learning model using the set of training data and transmits the trained first machine learning model to the controller. The controller generates an aggregate model by aggregating the trained first machine learning model with a plurality of other trained machine learning models that were transmitted to the controller by other computing devices. The aggregate model is sent to the wireless devices or maintained at the controller, and used by either the computing devices or the controller to generate outputs for the devices.

Claims20

Dependent claims

A method for training machine learning models in a wireless networking environment, the method comprising:

accessing, by a first wireless computing device, a set of training data that includes a set of input data points and a set of output data points, each output data point in the set of output data points defining a desired output from the first wireless computing device using at least one input data point in the set of input data points;
wherein the first wireless computing device is a vehicle;
training, by the first wireless computing device, a first machine learning model using the training data;
determining, by the first computing device, whether the first wireless computing device is driving or stationary;
responsive to a determination that the first wireless computing device is stationary, transmitting, by the first computing device, the first trained machine learning model to a remote controller;
responsive to a determination that the first wireless computing device is driving, transmitting, by the first computing device, the trained first machine learning model to a designated trusted device,
wherein the first wireless computing device has a first level of communication capability with the remote controller,
wherein the designated trusted device has a second level of communication capability with the remote controller that is greater than the first level of communication capability, and
wherein the first wireless computing device transmits the trained first machine learning model to the designated trusted device responsive to the second level of communication capability being greater than the first level of communication capability,
wherein the designated trusted device is associated with a same user as the first wireless computing device;
and
wherein the designated trusted device transmits the first trained machine learning model to the remote controller configured to aggregate the trained first machine learning model with one or more other trained machine learning models at the controller, wherein at least one of the other trained machine learning models at the controller are received via other wireless computing devices to generate an aggregate model;
and
applying, by the first wireless computing device, the aggregate model to generate an output by the first wireless computing device.

11 more elements

At least one computer-readable storage medium, excluding transitory signals and carrying instructions, which, when executed by at least one data processor of a system, cause the system to:

receive a first machine learning model trained by a first computing device;
receive a second machine learning model trained by a second computing device;
receive additional machine learning models from a plurality of computing devices other than the first computing device and the second computing device,
wherein the first computing device, the second computing device, and the plurality of computing devices comprise at least one vehicle configured to be either driving or stationary,
wherein, when at least one of a respective device of the first computing device, the second computing device, and the plurality of computing devices is stationary, the system receives a respective machine learning model of the first machine learning model, the second machine learning model, and the additional machine learning models directly from the respective device, and
wherein, when at least one of the respective device of the first computing device, the second computing device, and the plurality of computing devices is driving, the system receives the respective machine learning model of the first machine learning model, the second machine learning model, and the additional machine learning models via a designated trusted device;
receive real-time location information from the first computing device, the second computing device, and the plurality of computing devices,
wherein the real-time location information is determined using respective location sensors of the first computing device, the second computing device, and the plurality of computing devices,
wherein the first computing device and the plurality of computing devices are associated with a first application and, based on the location information, are located in a first geographical area,
wherein the first machine learning model and the additional machine learning models are each assigned a tag associating them with the first application and the first geographic area,
wherein the second computing device is associated with a second application different from the first application or is located in a second geographical area different from the first geographical area, and
wherein the second machine learning model is assigned a tag associating it with the second application or the second geographical area;
generate an aggregate machine learning model using the first machine learning model and the additional machine learning models,
wherein the first machine learning model and the additional machine learning models are selected for aggregation based on the tag associated with the first application and the first geographic area;
and
selectively transmit the aggregate machine learning model to the first computing device and the plurality of computing devices without transmitting the aggregate machine learning model to the second computing device.

13 more elements

Dependencies are inferred from the claim text.