Federated Learning Neural Graphical Models for Privacy

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current federated learning frameworks face challenges in handling multiple preselected outcome variables, dynamic neural network layers, and data privacy, particularly in domains like healthcare, where data remains proprietary and sensitive, leading to limitations in model accuracy and scalability.

Innovation Solution

The implementation of Neural Graphical Models within a federated learning framework that generates a global dependency graph from client-provided feature dependency graphs, allowing clients to train local models while maintaining data privacy, and aggregates knowledge to create a global model that represents the distribution over all features, enabling flexible modeling of complex distributions and multiple variables without sharing data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If federated learning is used to maintain data privacy across multiple clients, then data security is improved, but model accuracy and scalability deteriorate due to limitations in handling multiple outcome variables and data heterogeneity

Engineering Contradiction:
Improvedata privacyVSAvoidmodel accuracy
Core Design Contradiction:
ReliabilityVSMeasurement precision

Solution Approach 1:

The patent segments the learning process into local model training at each client and global model aggregation at the server. Each client trains on its own data locally, and the server aggregates these local models to create a global model. This segmentation allows data to remain private at client locations while still enabling collaborative learning across multiple clients, resolving the contradiction between data privacy and model accuracy.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces an intermediary global dependency graph that captures feature relationships across all clients without requiring direct data sharing. This intermediary structure serves as a mediator that enables accurate multi-variable prediction while maintaining data privacy, as it contains the necessary statistical dependencies without exposing actual client data.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Ease of manufacture

If traditional federated learning frameworks are used for single outcome variable prediction, then implementation simplicity is improved, but model versatility deteriorates when handling multiple preselected outcome variables

Engineering Contradiction:
Improveimplementation simplicityVSAvoidmodel versatility
Core Design Contradiction:
Ease of manufactureVSAdaptability or versatility

Solution Approach 1:

The patent implements a universal global dependency graph that can handle multiple outcome variables simultaneously. This graph structure serves multiple functions: it captures dependencies among all features across clients, supports predictions for any subset of outcome variables, and works with heterogeneous data types. This multi-functionality allows the same framework to handle single and multi-variable prediction tasks, improving versatility without sacrificing implementation simplicity.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Measurement precision

If global models aggregate all client data centrally, then model accuracy is improved, but data security deteriorates as proprietary information is exposed

Engineering Contradiction:
Improvemodel accuracyVSAvoiddata security
Core Design Contradiction:
Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The patent extracts only the necessary statistical dependencies and feature relationships from client data to construct the global dependency graph, rather than aggregating actual client data. Each client contributes local model parameters and dependency information, which are combined to form the global model structure. This extraction approach maintains model accuracy by capturing essential patterns while preventing exposure of proprietary client data, thus resolving the security-accuracy contradiction.

Inventive Principle:
Principle #2Taking out (Extraction)

4Reliability

If federated learning handles heterogeneous data across clients, then data privacy is maintained, but system complexity increases due to varying feature sets and data distributions

Engineering Contradiction:
Improvedata privacyVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent implements a dynamic global dependency graph that adapts to handle heterogeneous data structures across clients. The graph structure can dynamically include or exclude features based on their availability and relevance to different clients. This dynamic approach allows the system to accommodate varying feature sets and data distributions without requiring rigid preprocessing or feature alignment, thus managing complexity while maintaining data privacy.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentUS20250094824A1Federated learning with neural graphical models
Publication Date: 2025.03.20 MICROSOFT TECHNOLOGY LICENSING LLC
  • US20250094824A1 patent drawing
  • US20250094824A1 patent drawing
  • US20250094824A1 patent drawing

AI summary

The present disclosure relates to methods and systems that provide a federated learning framework using Neural Graphical Models. The federated learning framework combines the individual distributions learned by each client into a global model while keeping the data of each client private within each client's environment. The methods and systems allow for knowledge sharing among the clients without data sharing.