Multi-Model ML Application Updates for Wireless Signal Processing

Solution Overview

Problem

Existing wireless communication systems face inefficiencies in updating machine learning applications due to the large size of end-to-end models, which consume resources and impact performance when downloaded for condition changes.

Innovation Solution

A multi-model machine learning application is employed, comprising a static backbone model and task-specific models, allowing updates by replacing only the task-specific model, reducing the need for downloading entire applications.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If the entire end-to-end machine learning application is downloaded for condition changes, then the model can be updated to adapt to new conditions, but the time and resources consumed increase significantly

Engineering Contradiction:

Improvemodel adaptabilityVSAvoiddownload time

Core Design Contradiction:

Adaptability or versatilityVSLoss of time

Solution Approach 1:

The patent segments the end-to-end machine learning application into a backbone model and multiple task-specific models. The backbone model remains static and is not redownloaded, while only the relevant task-specific model is updated when conditions change. This segmentation allows selective downloading of model components, significantly reducing download time and resource consumption while maintaining the ability to adapt to different conditions by swapping task-specific models.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent extracts the task-specific model from the complete end-to-end application, allowing it to be updated independently. By separating the task-specific model from the backbone model, the system can download and apply only the necessary model component for condition changes, rather than downloading the entire application, thus reducing time and resource overhead.

Inventive Principle:

Principle #2Taking out (Extraction)

2Adaptability or versatility

If the entire end-to-end machine learning application is downloaded for condition changes, then the model can be updated to adapt to new conditions, but the network resources and device resources consumed increase

Engineering Contradiction:

Improvemodel adaptabilityVSAvoidnetwork resource consumption

Core Design Contradiction:

Adaptability or versatilityVSUse of energy by moving object

Solution Approach 1:

The patent segments the end-to-end machine learning application into a backbone model and multiple task-specific models. The backbone model remains static and is not redownloaded, while only the relevant task-specific model is updated when conditions change. This segmentation allows selective downloading of model components, significantly reducing download time and resource consumption while maintaining the ability to adapt to different conditions by swapping task-specific models.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent extracts the task-specific model from the complete end-to-end application, allowing it to be updated independently. By separating the task-specific model from the backbone model, the system can download and apply only the necessary model component for condition changes, rather than downloading the entire application, thus reducing time and resource overhead.

Inventive Principle:

Principle #2Taking out (Extraction)

3Adaptability or versatility

If the entire end-to-end machine learning application is downloaded for condition changes, then the model can be updated to adapt to new conditions, but the processing time and system performance are impacted

Engineering Contradiction:

Improvemodel adaptabilityVSAvoidupdate efficiency

Core Design Contradiction:

Adaptability or versatilityVSProductivity

Solution Approach 1:

The patent segments the end-to-end machine learning application into a backbone model and multiple task-specific models. The backbone model remains static and is not redownloaded, while only the relevant task-specific model is updated when conditions change. This segmentation allows selective downloading of model components, significantly reducing download time and resource consumption while maintaining the ability to adapt to different conditions by swapping task-specific models.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS12549971B2Configuring a multi-model machine learning application

Publication Date: 2026.02.10 QUALCOMM INC

AI summary

Methods, systems, and devices for wireless communications are described. A user equipment (UE) may receive a control message identifying a backbone model that is combinable with at least one task-specific model to generate a multi-model machine learning application. The UE may receive the backbone model and a first task-specific model identified by the control message. The UE use a multi-model machine learning application that is a combination of the backbone model and the first task-specific model to process one or more received signals to generate one or more outputs. The UE may communicate with a wireless device based on the one or more outputs.