Multiple AF Sessions with Group QoS for Federated Learning

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing wireless communication systems face challenges in supporting AI/ML model distribution and training over 5G networks, particularly in ensuring efficient and successful Federated Learning (FL) iterations by selecting the best group of user equipment (UEs) for AI/ML operations, while handling regular user traffic and maintaining required Quality of Service (QoS).

Innovation Solution

The system enhances 5G network functionality by exposing network status information to the AI/ML model provider, allowing for intelligent selection of UEs through multiple Application Function (AF) sessions with predefined QoS parameters, using Network Exposure Function (NEF) service operations to manage resource reservations and policy control for FL iterations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If multiple AF sessions are established for different UEs to support FL operations, then the capability to conduct AI/ML model training is improved, but the network complexity and resource management difficulty increase

Engineering Contradiction:

ImproveAI/ML model training capabilityVSAvoidnetwork complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system segments the FL operation into separate AF sessions for different UE groups, allowing independent QoS management for each session while maintaining overall coordination through the NEF and PCF. This enables the network to handle multiple AI/ML training operations simultaneously without overwhelming complexity.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The NEF acts as an intermediary between the AF and the core network, exposing network status information to the AF and managing the complexity of multiple AF sessions. The PCF serves as another intermediary, providing policy control for each AF session while abstracting the complexity from the AF.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Reliability

If QoS parameters are enforced for each AF session to ensure FL success, then the reliability of AI/ML operations is improved, but the difficulty of resource reservation and policy control increases

Engineering Contradiction:

ImproveFL operation success rateVSAvoidresource management complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The system performs preliminary action by establishing AF sessions with predefined QoS parameters before FL operations begin. The NEF pre-configures the sessions with appropriate QoS settings, and the PCF pre-establishes policy control rules, ensuring that when FL operations start, the necessary resources and quality guarantees are already in place.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system dynamically adjusts QoS parameters for different AF sessions based on network conditions and FL operation requirements. The NEF monitors network status and modifies QoS parameters accordingly, while the PCF updates policy control settings to maintain reliable FL operations under varying network conditions.

Inventive Principle:

Principle #35Parameter changes

3Productivity

If network status information is exposed to AF to enable intelligent UE selection, then the efficiency of FL iterations is improved, but the complexity of network exposure and information management increases

Engineering Contradiction:

ImproveFL iteration efficiencyVSAvoidnetwork exposure complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The NEF provides universal network status information exposure to AFs, serving multiple purposes: enabling intelligent UE selection for FL operations, monitoring network conditions, and facilitating resource optimization. This multi-functional approach increases FL iteration efficiency without proportionally increasing complexity.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The system implements feedback mechanisms where the NEF continuously monitors network status and provides updated information to the AF. This feedback loop enables the AF to make intelligent UE selection decisions based on current network conditions, improving FL iteration efficiency while the standardized feedback interface keeps complexity manageable.

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS12452146B2Supporting multiple application function sessions with required group quality of service (QoS) provided by machine learning model provider application function

Publication Date: 2025.10.21 INTEL CORP

AI summary

Embodiments herein provide techniques related to a NEF. An example technique includes identifying, from an AF that implements a ML model, a request related to reservation of resources for a plurality of AF sessions, wherein the plurality of AF sessions relate to respective UEs of a plurality of UEs, and wherein the request includes an indication of addresses of respective UEs of the plurality of UEs and a QoS parameter; transmitting, to a BSF of the cellular network, a discovery request related to discovery of respective PCFs that are serving the respective UEs of the plurality of UEs; identifying, based on the discovery request, a discovery response that includes indications of the PCFs; identifying, based on the indications of the PCFs and the QoS parameter, a result related to the reservation of resources; and transmitting, to the AF, an indication of the result. Other embodiments may be described and/or claimed.