UE Capability Reporting and Grouping for Federated Learning Delays

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing wireless communication systems face inefficiencies in federated learning processes due to varying machine learning capabilities and turnaround times among user equipment (UEs), leading to potential delays and bottlenecks in gradient updates, which affect the convergence time of the training procedure.

Innovation Solution

User equipment (UEs) report their machine learning processing capabilities and approximate turnaround times to the base station, enabling the base station to group UEs based on their capabilities, ensuring timely and efficient participation in federated learning rounds.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If UEs with varying machine learning capabilities participate in federated learning rounds without grouping, then all UEs can potentially contribute to gradient updates, but slower UEs cause delays and bottlenecks in the training procedure

Engineering Contradiction:
Improvefederated learning convergence speedVSAvoiddelay in gradient updates
Core Design Contradiction:
ProductivityVSLoss of time

Solution Approach 1:

The patent segments UEs into different groups based on their machine learning processing capabilities and turnaround times. The base station receives capability reports from UEs and forms capability groups, where each group contains UEs with similar processing speeds. This segmentation allows the system to schedule gradient collection rounds appropriately for each group, preventing slower UEs from becoming bottlenecks while maintaining overall system productivity.

Inventive Principle:
Principle #1Segmentation

2Reliability

If the base station waits for all UEs to complete gradient computations before proceeding, then no gradient updates are lost, but the convergence time increases due to slower UEs

Engineering Contradiction:
Improvecompletion of gradient updatesVSAvoidconvergence time
Core Design Contradiction:
ReliabilityVSDuration of action of moving object

Solution Approach 1:

The patent implements dynamic scheduling where the base station adjusts the number of rounds and timing of gradient collections based on reported UE capabilities. Rather than using a static wait-for-all approach, the system dynamically forms capability groups and schedules gradient collections to match each group's processing speed. This dynamic adaptation maintains reliability by ensuring gradients are collected when ready while minimizing convergence time by not waiting for slower UEs to complete their computations.

Inventive Principle:
Principle #15Dynamics

3Adaptability or versatility

If UEs report detailed machine learning capabilities, then the base station can optimize grouping and scheduling, but the signaling overhead and complexity increase

Engineering Contradiction:
ImproveUE capability matchingVSAvoidcapability reporting mechanism
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent changes the parameter representation of UE capabilities by having UEs report standardized capability metrics (such as processing speed categories and turnaround time estimates) rather than detailed hardware specifications. This parameter transformation enables the base station to perform effective capability-based grouping and scheduling while keeping the reporting mechanism simple and the signaling overhead manageable. The capability reports use quantized or categorized parameters that are easy to process and compare.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4226593B1User equipment (UE) capability report for machine learning applications
Publication Date: 2025.07.23 QUALCOMM INC
  • EP4226593B1 patent drawingFigure 1
  • EP4226593B1 patent drawingFigure 2
  • EP4226593B1 patent drawingFigure 3

AI summary

A method of wireless communication by a user equipment (UE) receives a machine learning model from a base station. The UE reports, to the base station, a machine learning processing capability. The UE also transmits, to the base station, gradient updates or weight updates to the machine learning model. A base station transmits a machine learning model to a number of UEs. The base station receives, from each of the number of UEs, a machine learning processing capability report. The base station groups a number of UEs in accordance with the machine learning processing capability reports, to receive gradient updates to the machine learning model.