Neural Utility Function Training for Interpretable Discrete Choice
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for designing utility functions in discrete choice models, such as manual design and neural network replacement, suffer from bias and reduced interpretability, which are not adequately addressed by current techniques.
Innovation Solution
A computer-implemented method that trains a neural network with parameters corresponding to the order and coefficient of explanatory variables in a utility function, allowing for the specification of a utility function that is interpretable for humans, using training data to optimize the network's structure and parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If a utility function is designed manually using expertise, then the utility function has high interpretability, but the design process is time-consuming and may introduce human bias
Solution Approach 1:
The patent applies preliminary action by pre-defining the functional form of the utility function with specific explanatory variables and their orders before the design process. This allows the machine learning model to directly estimate parameters within a constrained, interpretable framework, eliminating the need for time-consuming manual design while preserving interpretability through the predetermined functional structure.
2Measurement precision
If the entire part or part of a utility function is replaced with a Neural Network, then the modeling accuracy is improved, but the interpretability is reduced
Solution Approach 1:
The patent applies local quality by differentiating the treatment of different parts of the utility function. The functional form and explanatory variable structure are kept fixed and interpretable, while only the parameter estimation is performed using machine learning. This localized application of ML techniques to specific parameters maintains interpretability in the functional structure while improving accuracy through data-driven parameter optimization.
Solution Approach 2:
The patent applies parameter changes by using machine learning to estimate the parameters (coefficients and orders) of the utility function rather than specifying them manually. This allows the model to adapt parameters based on data while maintaining the overall functional form, thus improving accuracy without completely sacrificing interpretability since the functional structure remains visible and analyzable.
3Ease of manufacture
If manual design of utility function is performed, then the design process is simple and transparent, but human bias is introduced in the design
Solution Approach 1:
The patent applies self-service by enabling the utility function design process to be performed automatically by the machine learning model without human intervention. The system automatically selects explanatory variables, determines their orders, and estimates parameters based on the training data, eliminating human bias while maintaining design simplicity through automated workflows that require minimal human input.
Data Source
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.