Secure Logistic Regression Training With User-Tuned Convergence

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing secure computation techniques for logistic regression analysis face difficulties in adjusting analysis accuracy due to fixed convergence conditions and non-user-adjustable tuning parameters, leading to suboptimal model training.

Innovation Solution

A learning program and method that enables secure computation for logistic regression model training, allowing users to set and adjust tuning parameters such as learning rate, batch size, and reference values for update amounts, ensuring convergence based on user-specified criteria.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If fixed convergence conditions are used in secure computation training, then security is maintained, but analysis accuracy cannot be adjusted according to analysis subject

Engineering Contradiction:
Improveanalysis accuracyVSAvoidadjustability of convergence conditions
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent makes the convergence conditions dynamic by allowing users to specify tuning parameters such as the reference value for update amount, learning rate, and batch size. This enables the training process to adapt to different analysis subjects and requirements, resolving the contradiction between fixed security protocols and flexible accuracy adjustment.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent changes the parameters of the training process by introducing user-configurable tuning parameters that control convergence behavior. By allowing modification of parameters like learning rate, batch size, and convergence threshold, the system can optimize analysis accuracy for different scenarios while maintaining secure computation.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If tuning parameters are set in advance by developer, then security protocol is simplified, but user cannot adjust settings to optimize analysis accuracy

Engineering Contradiction:
Improveanalysis accuracyVSAvoiduser control over training parameters
Core Design Contradiction:
Measurement precisionVSEase of operation

Solution Approach 1:

The patent enables users to independently configure and adjust tuning parameters for the training process. Instead of relying on developer-preferred settings, users can specify their own parameters based on their analysis requirements, giving them control over the training process while maintaining the security protocol.

Inventive Principle:
Principle #25Self-service

3Productivity

If update amount threshold is fixed, then convergence determination is simplified, but training cannot be optimized for different analysis requirements

Engineering Contradiction:
Improvetraining efficiencyVSAvoidflexibility in convergence criteria
Core Design Contradiction:
ProductivityVSAdaptability or versatility

Solution Approach 1:

The patent makes the convergence criteria dynamic by allowing users to specify the reference value for update amount as a tuning parameter. This enables the system to adapt the convergence threshold to different analysis requirements, improving both training efficiency and flexibility simultaneously.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP4318441B1Learning program, learning method, and learning device
Publication Date: 2025.12.17 NTT DOCOMO BUSINESS INC
  • EP4318441B1 patent drawingFigure 1
  • EP4318441B1 patent drawingFigure 2
  • EP4318441B1 patent drawingFigure 3

AI summary

A calculating unit (233) of an analysis server (20) calculates, by secure computation, an objective variable by inputting an explanatory variable into a logistic regression model. An updating unit (234) updates, by secure computation, model parameters of the logistic regression model based on a result of calculation by the calculating unit (233). A determining unit (235) determines whether to end update of the model parameters based on whether an update amount of the model parameters by the updating unit (234) is smaller than a reference value that has been specified by a user. The reference value is input as an argument of a function.