Vertical Federated Learning Preparation for Private Sample Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing federated learning methods in 5G networks face challenges in training machine learning models without sharing raw data due to data privacy and security concerns, particularly in vertical federated learning scenarios where participants have different features for the same samples.
Innovation Solution
A method and device for vertical federated learning (VFL) that involves a VFL server transmitting preparation requests to VFL clients, including a machine learning preparation flag, and checking for ML model training requirements, using services like Nnwdaf_MLModelTraining_Subscribe or Nnef_VFLPreparation_Subscribe, to align samples and train models without data exchange.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If raw data is shared across network functions for machine learning model training, then model training effectiveness is improved, but data privacy and security are compromised
Solution Approach 1:
The patent segments the data processing workflow into distinct phases: data collection at source network functions, local preprocessing and feature extraction, and centralized model training at the NWDAF. This segmentation allows each entity to process only the data it needs to retain, eliminating the need to share raw data while maintaining training effectiveness through coordinated intermediate result exchange.
Solution Approach 2:
The patent introduces an intermediary data processing layer that transforms raw data into intermediate representations (features, statistics, or model updates) before transmission to the central training entity. This intermediary processing acts as a privacy-preserving barrier, allowing model training to proceed on derived data without exposing sensitive raw information across network boundaries.
2Object-affected harmful factors
If federated learning is implemented without data sharing, then data privacy is preserved, but model training complexity increases due to coordination overhead
Solution Approach 1:
The patent defines a universal preparation request message structure that can accommodate multiple data types and processing scenarios. The standardized message format includes fields for various intermediate result representations (features, statistics, gradients) and supports different service operations (Nnwdaf_MLModelTraining_Subscribe, Nnef_VFLPreparation_Subscribe), reducing coordination complexity through a unified interface that handles diverse federated learning workflows.
Solution Approach 2:
The patent changes the state of data from raw form to processed intermediate representations through standardized transformation parameters defined in the preparation request. By establishing conventional parameter transformations (feature extraction methods, aggregation functions, encoding schemes), the system reduces coordination complexity through predictable, standardized data state transitions that simplify inter-entity communication.
3Object-affected harmful factors
If vertical federated learning is performed with different feature sets, then data privacy is maintained, but sample alignment difficulty increases
Solution Approach 1:
The patent implements self-service sample alignment mechanisms where each network function independently processes its local data through standardized preparation operations defined in the request message. Each entity performs feature extraction, filtering, or transformation locally based on its available features, and the standardized message structure ensures that intermediate results are automatically compatible for aggregation, eliminating complex cross-entity alignment procedures.
Solution Approach 2:
The patent performs preliminary data preparation and feature extraction at each participating network function before central aggregation. The preparation request message specifies preprocessing operations to be executed locally, ensuring that data is transformed into a standardized intermediate format in advance. This preliminary action resolves sample alignment issues beforehand, allowing the central NWDAF to directly aggregate results without complex alignment procedures.
Data Source
AI summary
Disclosed are a method of supporting federated learning based on non-sharing of original data and a device for performing the same. An operating method of a vertical federated learning (VFL) server according to an embodiment may include transmitting a VFL preparation request to a VFL client, receiving a response to the VFL preparation request from the VFL client, and performing VFL with the VFL client, wherein the VFL preparation request may include a machine learning (ML) preparation flag.