Federated Learning Model Parameter Concealment

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

There is a risk that malicious users can infer chemical compound data used for federated learning by obtaining the parameters of machine learning models.

Innovation Solution

A computation system and method that includes concealment means to generate models from chemical compound data and conceal the model parameters, and secure computation means to integrate the models using the concealed parameters.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If machine learning models are trained using chemical compound data from multiple organizations, then the utility and accuracy of the models are improved, but the risk of data leakage and inference increases

Engineering Contradiction:

Improvemodel accuracyVSAvoiddata leakage risk

Core Design Contradiction:

Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The patent segments the training process into local model training at each organization and centralized model integration at the server. Each organization trains its own local model independently, and only the integrated model parameters are combined at the server, preventing direct access to raw data while maintaining model accuracy through federated learning

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces concealed parameters as an intermediary between the local model parameters and the integrated model. The concealment unit transforms local parameters into concealed form before transmission, and the integration unit integrates these concealed parameters to generate the final model, thereby preventing direct inference of original data while preserving model utility

Inventive Principle:

Principle #24Intermediary (Mediator)

2Reliability

If secure computation methods are used to conceal parameters during federated learning, then data security is improved, but computation time increases

Engineering Contradiction:

Improvedata securityVSAvoidcomputation time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent applies partial concealment rather than complete encryption of all model parameters. The concealment unit selectively conceals parameters that could lead to data inference while maintaining other parameters in a more computationally efficient format, achieving security without excessive computation time

Inventive Principle:

Principle #16Partial or excessive action

Solution Approach 2:

The patent transforms model parameters into concealed parameters through a mathematical transformation that maintains the utility of the parameters for model integration while reducing the risk of data inference. This parameter transformation enables secure computation with realistic computation times by changing the representation form rather than using heavy encryption

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS20250182861A1Computation system and computation method

Publication Date: 2025.06.05 NEC CORP

AI summary

To provide a computation system and a computation method for reducing the risk of inferring chemical compound data used for federated learning. A computation system includes: a concealment unit configured to perform a first processing of generating a model from a set of chemical compound data stored in each of a plurality of client terminals, and then concealing parameters of the generated model; and a secure computation unit configured to perform secure computation for integrating the generated models using the concealed parameters.