# US 2023/0274181 A1 · Federated learning model splitting and combining

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

- Number: US 2023/0274181 A1
- Document ID: US20230274181A1
- Status: Published application
- Kind: A1
- Application number: 17/652,909
- Filed: 2022-02-28
- Published: 2023-08-31
- Earliest claimed: 2022-02-28
- Assignees: QUALCOMM Incorporated
- Applicants: QUALCOMM Incorporated
- Inventors: Tianyang BAI; Hua WANG; Junyi LI
- Web page: https://patentgrep.com/patent/US20230274181A1

## CPC

- G06N 20/00: Machine learning
- G06N 3/045: Combinations of networks
- H04W 24/02: Arrangements for optimising operational condition
- G06N 3/098: Distributed learning, e.g. federated learning
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- H04W: WIRELESS COMMUNICATION NETWORKS (broadcast communication H04H; communication systems using wireless links for non-selective communication, e.g. wireless extensions H04M1/72)

## Abstract

Various aspects of the present disclosure generally relate to wireless communication. In some aspects, a network node may receive, from a first user equipment (UE) and a second UE, one or more parameter updates associated with a first model for federated learning. The network node may transmit, to the first UE, a second model for federated learning that is based at least in part on the first model. The network node may transmit, to the second UE, a third model for federated learning that is based at least in part on the first model. Numerous other aspects are described.

## Claims

### Claim 1

An apparatus for wireless communication at a network node, comprising:
a memory; and
one or more processors coupled to the memory, the one or more processors configured to:
receive, from a first user equipment (UE) and a second UE, one or more parameter updates associated with a first model for federated learning;
transmit, to the first UE, a second model for federated learning that is based at least in part on the first model; and
transmit, to the second UE, a third model for federated learning that is based at least in part on the first model.

### Claim 2 (depends on 1)

The apparatus of claim 1, wherein the one or more processors are further configured to generate the second model and the third model based at least in part on the first model.

### Claim 3 (depends on 2)

The apparatus of claim 2, wherein the one or more processors, to generate the second model and the third model based at least in part on the first model, are configured to split the first model into the second model and the third model.

### Claim 4 (depends on 1)

The apparatus of claim 1, wherein the one or more processors, to receive the one or more parameter updates, are configured to receive a first gradient from the first UE and receive a second gradient from the second UE.

### Claim 5 (depends on 1)

The apparatus of claim 1, wherein the one or more processors are further configured to:
generate a group of parameter updates that includes a parameter update from the first UE and a parameter update from the second UE, and
generate the second model and the third model based at least in part on the group of parameter updates.

### Claim 6 (depends on 5)

The apparatus of claim 5, wherein the one or more processors, to generate the second model and the third model based at least in part on the group of parameter updates, are configured to:
determine an average parameter update based at least in part on the group of parameter updates;
generate the second model based at least in part on a difference between the parameter update from the first UE and the average parameter update; and
generate the third model based at least in part on a difference between the parameter update from the second UE and the average parameter update.

### Claim 7 (depends on 1)

The apparatus of claim 1, wherein the one or more processors are further configured to transmit information that enables the first UE and the second UE to respectively generate one or more error metrics based at least in part on the first model.

### Claim 8 (depends on 7)

The apparatus of claim 7, wherein the one or more processors, to receive the one or more parameter updates, are configured to receive a first error metric from the first UE and receive a second error metric from the second UE.

### Claim 9 (depends on 7)

The apparatus of claim 7, wherein the information that enables the first UE and the second UE to generate the one or more error metrics includes a threshold error metric.

### Claim 10 (depends on 1)

The apparatus of claim 1, wherein the one or more processors are further configured to maintain first training information associated with the second model and second training information associated with the third model.

### Claim 11 (depends on 1)

The apparatus of claim 1, wherein the first model comprises a first layer and a second layer, and wherein receiving the one or more parameter updates comprises receiving one or more parameter updates associated with the first layer.

### Claim 12 (depends on 11)

The apparatus of claim 11, wherein the second model and the third model are based at least in part on the first layer of the first model.

### Claim 13

An apparatus for wireless communication at a network node, comprising:
a memory; and
one or more processors coupled to the memory, the one or more processors configured to:
receive, from a first user equipment (UE), a first parameter update associated with a first model for federated learning;
receive, from a second UE, a second parameter update associated with a second model for federated learning; and
transmit, to the first UE and the second UE, a third model for federated learning that is based at least in part on the first model and the second model.

### Claim 14 (depends on 13)

The apparatus of claim 13, wherein the one or more processors are further configured to generate the third model based at least in part on the first model and the second model.

### Claim 15 (depends on 14)

The apparatus of claim 14, wherein the one or more processors, to generate the third model based at least in part on the first model and the second model, are configured to combine the first model and the second model.

### Claim 16 (depends on 13)

The apparatus of claim 13, wherein the first UE is configured with the first model and the second model, and the second UE is configured with the first model and the second model.

### Claim 17 (depends on 16)

The apparatus of claim 16, wherein the first model is the primary model for the first UE, and the second model is the primary model for the second UE.

### Claim 18 (depends on 16)

The apparatus of claim 16, wherein the second model is the primary model for the first UE, and the first model is the primary model for the second UE.

### Claim 19 (depends on 18)

The apparatus of claim 18, wherein the one or more processors are further configured to:
transmit, to the first UE, information for determining the first parameter update based at least in part on the first model; and
transmit, to the second UE, information for determining the second parameter update based at least in part on the second model.

### Claim 20 (depends on 19)

The apparatus of claim 19, wherein the information for determining the first parameter update comprises information for determining the first parameter update based at least in part on a first average parameter update, and the information for determining the second parameter update comprises information for determining the second parameter update based at least in part on a second average parameter update.

### Claim 21 (depends on 19)

The apparatus of claim 19, wherein the first parameter update includes a first error metric, and the second parameter update includes a second error metric.

### Claim 22 (depends on 13)

The apparatus of claim 13, wherein the one or more processors are further configured to maintain, for the first UE and the second UE, training information associated with the third model.

### Claim 23 (depends on 13)

The apparatus of claim 13, wherein the first model comprises a first layer of the first model and a second layer of the first model, and the second model comprises a first layer of the second model and a second layer of the second model, wherein receiving the first parameter update comprises receiving a first parameter update associated with the second layer of the first model, and wherein receiving the second parameter update comprises receiving a second parameter update associated with the second layer of the second model.

### Claim 24 (depends on 23)

The apparatus of claim 23, wherein the third model is based at least in part on the second layer of the first model and the second layer of the second model.

### Claim 25

A method of wireless communication performed by a network node, comprising:
receiving, from a first user equipment (UE) and a second UE, one or more parameter updates associated with a first model for federated learning;
transmitting, to the first UE, a second model for federated learning that is based at least in part on the first model; and
transmitting, to the second UE, a third model for federated learning that is based at least in part on the first model.

### Claim 26 (depends on 25)

The method of claim 25, further comprising generating the second model and the third model based at least in part on the first model.

### Claim 27 (depends on 26)

The method of claim 26, wherein generating the second model and the third model based at least in part on the first model comprises splitting the first model into the second model and the third model.

### Claim 28

A method of wireless communication performed by a network node, comprising:
receiving, from a first user equipment (UE), a first parameter update associated with a first model for federated learning;
receiving, from a second UE, a second parameter update associated with a second model for federated learning; and
transmitting, to the first UE and the second UE, a third model for federated learning that is based at least in part on the first model and the second model.

### Claim 29 (depends on 28)

The method of claim 28, further comprising generating the third model based at least in part on the first model and the second model.

### Claim 30 (depends on 29)

The method of claim 29, wherein generating the third model based at least in part on the first model and the second model comprises combining the first model and the second model.