# US 2024/0174254 A1 · Systems and methods for communication-aware federated learning

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2024/0174254 A1
- Document ID: US20240174254A1
- Status: Published application
- Kind: A1
- Application number: 17/994,850
- Filed: 2022-11-28
- Published: 2024-05-30
- Earliest claimed: 2022-11-28
- Assignees: Toyota Motor Engineering & Manufacturing North America, Inc.; Toyota Jidosha Kabushiki Kaisha
- Applicants: Toyota Motor Engineering & Manufacturing North America, Inc.
- Inventors: Yitao Chen; Dawei Chen; Haoxin Wang; Kyungtae Han
- Web page: https://patentgrep.com/patent/US20240174254A1

## CPC

- B60W 60/001: {Planning or execution of driving tasks}
- G06F 18/214: Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- B60W 2420/403: Image sensing, e.g. optical camera
- B60W: CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

A system for communication-aware federated learning includes a server and edge nodes. Each of the edge nodes trains a machine learning model using first local data obtained by sensors of corresponding edge node. Each of the edge nodes obtains network bandwidth for a channel between corresponding edge node and the server. One or more of the edge nodes determines a level of compression based on the bandwidth for the channel, compresses the trained machine leaning model based on the determined level of compression, and transmits the compressed trained machine learning model to the server. The server decompresses the compressed trained machine learning models and aggregates the decompressed trained machine learning models to obtain the aggregated machine learning model, and transmits the aggregated machine learning model to each of the edge nodes. Each of the edge nodes receives the aggregated machine learning model from the server.

## Claims

### Claim 1

A vehicle comprising:
a controller programmed to:
train a machine learning model using first local data;
obtain a network bandwidth for a channel between the vehicle and a server;
determine a level of compression based on the network bandwidth for the channel;
compress the trained machine leaning model based on the determined level of compression;
transmit the compressed trained machine learning model to the server;
receive an aggregated machine learning model from the server; and
control the vehicle to drive autonomously based on the aggregated machine learning model.

### Claim 2 (depends on 1)

The vehicle according to claim 1, wherein the controller is further programmed to:
train the aggregated machine learning model using second local data;
detect a change in the network bandwidth for the channel;
determine another level of compression based on the changed network bandwidth;
compress the trained aggregated machine learning model based on the another level of compression; and
transmit the compressed trained aggregated machine learning model to the server.

### Claim 3 (depends on 1)

The vehicle according to claim 1, wherein the level of compression is inversely correlated to the network bandwidth.

### Claim 4 (depends on 1)

The vehicle according to claim 1, wherein the controller is further programmed to:
compress the trained machine leaning model by quantizing parameters of the trained machine leaning model.

### Claim 5 (depends on 1)

The vehicle according to claim 1, wherein the controller is programmed to:
compress the trained machine leaning model based on knowledge distillation of the trained machine leaning model.

### Claim 6 (depends on 1)

The vehicle according to claim 1, wherein the controller is programmed to:
transmit, to the server, information about a compression scheme used to compress the trained machine leaning model.

### Claim 7 (depends on 1)

The vehicle according to claim 1, further comprising:
an imaging sensor configured to capture the first local data.

### Claim 8 (depends on 1)

The vehicle according to claim 1, wherein the controller is a graphics processing unit.

### Claim 9 (depends on 1)

The vehicle according to claim 1, wherein the controller is programmed to:
determine whether the trained machine leaning model can be completely uploaded to the server within a designated time window based on the obtained network bandwidth; and
compress the trained machine leaning model based on the determined level of compression in response to determining that the trained machine leaning model cannot be completely uploaded to the server within a designated time window.

### Claim 10 (depends on 1)

The vehicle according to claim 1, wherein the controller is programmed to operate one or more actuators of the vehicle to keep the vehicle within lane boundaries using the aggregated machine learning model.

### Claim 11

A method for communication-aware federated learning, the method comprising:
training a machine learning model using first local data;
obtaining a network bandwidth for a channel between a vehicle and a server;
determining a level of compression based on the network bandwidth for the channel;
compressing the trained machine leaning model based on the determined level of compression;
transmitting the compressed trained machine learning model to the server;
receiving an aggregated machine learning model from the server; and
controlling the vehicle to drive autonomously based on the aggregated machine learning model.

### Claim 12 (depends on 11)

The method of claim 11, further comprising:
training the aggregated machine learning model using second local data;
detecting a change in the network bandwidth for the channel;
determining another level of compression based on the changed network bandwidth;
compressing the trained aggregated machine learning model based on the another level of compression; and
transmitting the compressed trained aggregated machine learning model to the server.

### Claim 13 (depends on 11)

The method of claim 11, wherein the level of compression is inversely correlated to the network bandwidth.

### Claim 14 (depends on 11)

The method of claim 11, further comprising:
transmitting, to the server, information about a compression scheme used to compress the trained machine leaning model.

### Claim 15 (depends on 11)

The method of claim 11, further comprising:
determining whether the trained machine leaning model can be completely uploaded to the server within a designated time window based on the obtained network bandwidth; and
compressing the trained machine leaning model based on the determined level of compression in response to determining that the trained machine leaning model cannot be completely uploaded to the server within a designated time window.

### Claim 16

A system for communication-aware federated learning, the system comprising:
a server; and
a plurality of vehicles, each of the vehicles comprising a controller programmed to:
train a machine learning model using first local data;
obtain a network bandwidth for a channel between the vehicle and the server;
determine a level of compression based on the bandwidth for the channel;
compress the trained machine leaning model based on the determined level of compression;
transmit the compressed trained machine learning model to the server;
receive an aggregated machine learning model from the server; and
control the vehicle to drive autonomously based on the aggregated machine learning model,
wherein the server generates the aggregated machine learning model based on the compressed trained machine learning models from the plurality of vehicles.

### Claim 17 (depends on 16)

The system of claim 16, wherein the server comprises a controller programmed to decompress the compressed trained machine learning models.

### Claim 18 (depends on 16)

The system of claim 16, wherein the controller is further programmed to:
train the aggregated machine learning model using second local data;
detect a change in the network bandwidth for the channel;
determine another level of compression based on the changed network bandwidth;
compress the trained aggregated machine learning model based on the another level of compression; and
transmit the compressed trained aggregated machine learning model to the server.

### Claim 19 (depends on 16)

The system of claim 16, wherein the level of compression is inversely correlated to the network bandwidth.

### Claim 20 (depends on 16)

The system of claim 16, wherein the controller is programmed to:
determine whether the trained machine leaning model can be completely uploaded to the server within a designated time window based on the obtained network bandwidth; and
compress the trained machine leaning model based on the determined level of compression in response to determining that the trained machine leaning model cannot be completely uploaded to the server within a designated time window.
