# US 2023/0276480 A1 · Multi-head machine learning model for estimating a plurality of parameter values

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

- Number: US 2023/0276480 A1
- Document ID: US20230276480A1
- Status: Published application
- Kind: A1
- Application number: 17/652,854
- Filed: 2022-02-28
- Published: 2023-08-31
- Earliest claimed: 2022-02-28
- Assignees: QUALCOMM Incorporated
- Applicants: QUALCOMM Incorporated
- Inventors: Mohamed Fouad Ahmed MARZBAN; Wooseok NAM; Tao LUO; Taesang YOO
- Web page: https://patentgrep.com/patent/US20230276480A1

## CPC

- G06N 3/045: Combinations of networks
- H04W 72/54: based on quality criteria
- G06N 3/08: Learning methods
- G06N 3/09: Supervised learning
- G06N 3/096: Transfer learning
- G06N 3/098: Distributed learning, e.g. federated learning
- H04W 24/02: Arrangements for optimising operational condition
- 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 first network node may receive a signal associated with a set of resources. The first network node may determine, using a multi-head machine learning model having a body module and a plurality of head modules, a plurality of estimated parameter values of a plurality of parameters corresponding to the set of resources, wherein the body module extracts a set of common features based at least in part on a set of model inputs corresponding to the set of resources, and wherein the plurality of head modules generate the plurality of estimated parameter values based at least in part on the set of common features. The first network node may perform a wireless communication operation based at least in part on the plurality of estimated parameter values. Numerous other aspects are described.

## Claims

### Claim 1

A first network node for wireless communication, comprising:
a memory; and
one or more processors, coupled to the memory, configured to:
receive a signal associated with a set of resources;
determine, using a multi-head machine learning model having a body module and a plurality of head modules, a plurality of estimated parameter values of a plurality of parameters corresponding to the set of resources, wherein the body module extracts a set of common features based at least in part on a set of model inputs corresponding to the set of resources, and wherein the plurality of head modules generate the plurality of estimated parameter values based at least in part on the set of common features; and
perform a wireless communication operation based at least in part on the plurality of estimated parameter values.

### Claim 2 (depends on 1)

The first network node of claim 1, wherein the set of resources comprises at least one of a time domain resource, a frequency domain resource, an angular domain resource, or a spatial domain resource.

### Claim 3 (depends on 1)

The first network node of claim 1, wherein each head module of the plurality of head modules generates only one estimated parameter value of the plurality of estimated parameter values.

### Claim 4 (depends on 1)

The first network node of claim 1, wherein the plurality of estimated parameter values comprises at least one of an estimated parameter value associated with a current resource of the set of resources or a predicted parameter value associated with a future resource of the set of resources.

### Claim 5 (depends on 1)

The first network node of claim 1, wherein a head module of the plurality of head modules generates an estimated parameter value of the plurality of estimated parameter values based at least in part on minimizing a loss function associated with a parameter of the plurality of parameters.

### Claim 6 (depends on 5)

The first network node of claim 5, wherein the one or more processors are further configured to train the multi-head machine learning model based at least in part on a total loss function associated with the multi-head machine learning model, wherein the total loss function comprises a weighted sum of the loss function and at least one additional loss function associated with at least one additional parameter of the plurality of parameters.

### Claim 7 (depends on 6)

The first network node of claim 6, wherein the one or more processors are further configured to receive a machine learning configuration that configures a first weight associated with the loss function and a second weight associated with the at least one additional loss function.

### Claim 8 (depends on 1)

The first network node of claim 1, wherein the multi-head machine learning model comprises a neural network.

### Claim 9 (depends on 8)

The first network node of claim 8, wherein a head module of the plurality of head modules comprises a single machine learning neuron, a single layer of multiple machine learning neurons, or a plurality of layers, each layer including at least one machine learning neuron.

### Claim 10 (depends on 1)

The first network node of claim 1, wherein the plurality of estimated parameter values comprises a plurality of estimated interference distributions associated with a plurality of resources of the set of resources, wherein the set of common features comprises a set of interference features common to the plurality of resources.

### Claim 11 (depends on 1)

The first network node of claim 1, wherein the plurality of estimated parameter values comprises:
a plurality of values associated with a set of estimated channel state information parameter values,
at least one estimated channel state information parameter value and at least one estimated interference parameter value, or
a plurality of predicted reference signal received power values corresponding to a plurality of beams.

### Claim 12 (depends on 1)

The first network node of claim 1, wherein the set of model inputs are based at least in part on at least one reference signal.

### Claim 13 (depends on 12)

The first network node of claim 12, wherein the plurality of estimated parameter values comprises a first estimated parameter value corresponding to a first parameter and a second estimated parameter value corresponding to a second parameter that is different than the first parameter, wherein the first estimated parameter value is based at least in part on the at least one reference signal, and wherein the second estimated parameter value is based at least in part on the at least one reference signal.

### Claim 14 (depends on 12)

The first network node of claim 12, wherein the plurality of estimated parameter values comprises a first estimated parameter value corresponding to a parameter in connection with a first resource and a second estimated parameter value corresponding to the parameter in connection with a second resource that is different than the first resource, wherein the first estimated parameter value is based at least in part on the at least one reference signal, and wherein the second estimated parameter value is based at least in part on the at least one reference signal.

### Claim 15 (depends on 1)

The first network node of claim 1, wherein the one or more processors are further configured to receive a machine learning configuration that configures at least one of:
a head module of the plurality of head modules,
a quantity of head modules associated with the multi-head machine learning model, or
a parameter, of the plurality of parameters, corresponding to a head module of the plurality of head modules.

### Claim 16 (depends on 1)

The first network node of claim 1, wherein the one or more processors are further configured to receive a machine learning configuration that configures a resource granularity associated with the set of resources at a head module of the plurality of head modules.

### Claim 17 (depends on 1)

The first network node of claim 1, wherein the multi-head machine learning model comprises at least one additional head module, wherein the one or more processors are further configured to receive a head module deactivation indication associated with the at least one additional head module, and wherein the plurality of head modules excludes the at least one additional head module.

### Claim 18 (depends on 1)

The first network node of claim 1, wherein the one or more processors are further configured to:
transmit, to a second network node, a capability report that indicates a capability associated with the multi-head machine learning model; and
receive, from the second network node and based at least in part on the capability report, a machine learning configuration associated with the multi-head machine learning model.

### Claim 19 (depends on 18)

The first network node of claim 18, wherein the capability report indicates at least one of:
implementation of the multi-head machine learning model, or
a maximum quantity of head modules associated with the multi-head machine learning model.

### Claim 20 (depends on 1)

The first network node of claim 1, wherein the one or more processors are further configured to transmit an output report that indicates at least one of the plurality of estimated parameter values.

### Claim 21

A first network node for wireless communication, comprising:
a memory; and
one or more processors, coupled to the memory, configured to:
transmit, to a second network node, a machine learning configuration associated with a multi-head machine learning model having a body module and a plurality of head modules, wherein the multi-head machine learning model is configured to determine a plurality of estimated parameter values of a plurality of parameters corresponding to a set of resources, wherein the body module is configured to extract a set of common features based at least in part on a set of model inputs corresponding to the set of resources, and wherein the plurality of head modules are configured to generate the plurality of estimated parameter values based at least in part on the set of common features; and
receive, from the second network node, an output report that indicates at least one of the plurality of estimated parameter values.

### Claim 22 (depends on 21)

The first network node of claim 21, wherein the set of resources comprises at least one of a time domain resource, a frequency domain resource, an angular domain resource, or a spatial domain resource.

### Claim 23 (depends on 21)

The first network node of claim 21, wherein the multi-head machine learning model comprises a neural network.

### Claim 24 (depends on 21)

The first network node of claim 21, wherein the machine learning configuration further configures at least one of:
a head module of the plurality of head modules,
a quantity of head modules associated with the multi-head machine learning model,
a parameter, of the plurality of parameters, corresponding to a head module of the plurality of head modules,
a resource granularity associated with the set of resources at a head module of the plurality of head modules, or
a weight associated with a loss function associated with the plurality of parameters.

### Claim 25 (depends on 21)

The first network node of claim 21, wherein the multi-head machine learning model comprises at least one additional head module, and wherein the one or more processors are further configured to transmit a head module deactivation indication associated with the at least one additional head module, and wherein the plurality of head modules excludes the at least one additional head module.

### Claim 26 (depends on 21)

The first network node of claim 21, wherein the one or more processors are further configured to receive, from the second network node, a capability report that indicates a capability associated with the multi-head machine learning model, and wherein the one or more processors, to transmit the machine learning configuration, are configured to transmit the machine learning configuration based at least in part on the capability report.

### Claim 27

A method of wireless communication performed by a first network node, comprising:
receiving a signal associated with a set of resources;
determining, using a multi-head machine learning model having a body module and a plurality of head modules, a plurality of estimated parameter values of a plurality of parameters corresponding to the set of resources, wherein the body module extracts a set of common features based at least in part on a set of model inputs corresponding to the set of resources, and wherein the plurality of head modules generate the plurality of estimated parameter values based at least in part on the set of common features; and
performing a wireless communication operation based at least in part on the plurality of estimated parameter values.

### Claim 28 (depends on 27)

The method of claim 27, further comprising receiving a machine learning configuration that configures at least one of:
a head module of the plurality of head modules,
a quantity of head modules associated with the multi-head machine learning model,
a parameter, of the plurality of parameters, corresponding to a head module of the plurality of head modules, or
a resource granularity associated with the set of resources at a head module of the plurality of head modules.

### Claim 29

A method of wireless communication performed by a first network node, comprising:
transmitting, to a second network node, a machine learning configuration associated with a multi-head machine learning model having a body module and a plurality of head modules, wherein the multi-head machine learning model is configured to determine a plurality of estimated parameter values of a plurality of parameters corresponding to a set of resources, wherein the body module is configured to extract a set of common features based at least in part on a set of model inputs corresponding to the set of resources, and wherein the plurality of head modules are configured to generate the plurality of estimated parameter values based at least in part on the set of common features; and
receiving, from the second network node, an output report that indicates at least one of the plurality of estimated parameter values.

### Claim 30 (depends on 29)

The method of claim 29, wherein the multi-head machine learning model comprises at least one additional head module, the method further comprising transmitting a head module deactivation indication associated with the at least one additional head module, and wherein the plurality of head modules excludes the at least one additional head module.
