Neural Utility Function Learning for Interpretable Choice Models

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for designing utility functions in discrete choice models either introduce bias due to human intervention or compromise interpretability when using neural networks, leading to a loss of transparency.

Innovation Solution

A computer-implemented method trains a neural network with parameters corresponding to the structure and coefficients of explanatory variables in a discrete choice model, allowing for the specification of an interpretable utility function.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of information

If a utility function is designed manually using expert knowledge, then the utility function has high interpretability, but bias may be introduced in the explanatory variable or parameter

Engineering Contradiction:
ImproveinterpretabilityVSAvoidbias
Core Design Contradiction:
Loss of informationVSReliability

Solution Approach 1:

The patent introduces an intermediary process between manual design and final utility function by using machine learning models to learn from data while maintaining interpretability constraints. This intermediary approach allows the system to learn from training data without complete manual specification, reducing bias while preserving interpretability through structured parameter learning.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If the entire part or part of a utility function is replaced with a Neural Network, then the possibility of bias is reduced, but the interpretability is degraded

Engineering Contradiction:
Improvebias reductionVSAvoidinterpretability
Core Design Contradiction:
ReliabilityVSLoss of information

Solution Approach 1:

The patent applies local quality by selectively replacing only specific components of the utility function with neural network elements while maintaining the interpretability of other parts. Instead of complete replacement, the approach uses localized neural network integration that preserves the explanatory structure where needed while leveraging machine learning where pattern recognition is most beneficial.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent creates a composite approach by combining traditional discrete choice model elements with neural network components. This hybrid structure integrates the interpretability of parametric models with the predictive power of neural networks, forming a composite utility function that leverages the strengths of both approaches while mitigating their respective weaknesses.

Inventive Principle:
Principle #40Composite materials

3Ease of operation

If a discrete choice model is used to model human behavior, then the logic is understandable and the utility function has high interpretability, but manual design may introduce human bias

Engineering Contradiction:
ImproveunderstandabilityVSAvoidhuman bias
Core Design Contradiction:
Ease of operationVSReliability

Solution Approach 1:

The patent enables the system to self-correct bias by using data-driven learning instead of relying solely on human expert judgment. The machine learning models learn patterns directly from training data, allowing the system to serve itself in identifying relationships without human intervention, thereby reducing the introduction of human bias while maintaining the structured framework of discrete choice models.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP4700654A1Method for machine learning and machine learning program
Publication Date: 2026.02.25 FUJITSU LTD
  • EP4700654A1 patent drawingFigure 1
  • EP4700654A1 patent drawingFigure 2
  • EP4700654A1 patent drawingFigure 3

AI summary

A method for machine learning includes training a neural network including parameters, at least some of the parameters having a structure corresponding to an order and a coefficient of an explanatory variable of a utility function of a discrete choice model, using training data including a value of the explanatory variable and a choice result; and specifying the utility function in the neural network after being subjected to the training.