Constraint-Aware ML Training for Feasible Improvement Proposals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models struggle to provide improvement proposals that balance prediction accuracy and feasibility, often resulting in unfeasible suggestions like reducing age to achieve desired outcomes.
Innovation Solution
A training program that calculates an evaluation index based on prediction accuracy and feasibility, using constraint condition data to train the model, dividing training data into regions, and recursively training decision trees to optimize both accuracy and feasibility.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the machine learning model is trained to maximize prediction accuracy, then the prediction accuracy is improved, but the feasibility of improvement proposals deteriorates
Solution Approach 1:
The patent introduces a dual-objective training approach that changes the optimization parameters from single-purpose accuracy maximization to a combined objective function incorporating both prediction accuracy and proposal feasibility. The model learns to balance these two competing objectives through joint optimization during training, allowing it to generate proposals that are both accurate and feasible simultaneously.
Solution Approach 2:
The patent segments the training process into two distinct but integrated components: one optimizing for prediction accuracy and another optimizing for proposal feasibility. By dividing the objective function into these separate components that are then combined, the model can independently learn each aspect while maintaining their balance through the composite loss function.
2Measurement precision
If the machine learning model provides improvement proposals without considering constraints, then the prediction accuracy is improved, but the feasibility of proposals deteriorates
Solution Approach 1:
The patent incorporates feedback mechanisms where the model's proposed improvements are evaluated against constraint conditions during training. The feasibility assessment provides feedback signals that guide the model to adjust its proposals, creating a closed-loop learning process that continuously improves proposal reliability while maintaining prediction accuracy.
Solution Approach 2:
The patent performs preliminary evaluation of proposal feasibility against constraint conditions before finalizing the improvement suggestions. By pre-assessing whether proposed changes satisfy business rules and constraints, the model可以避免 generating infeasible proposals and can focus on finding solutions that are both accurate and actionable.
3Productivity
If the model focuses on generating any improvement proposal, then the productivity is improved, but the quality of feasible proposals deteriorates
Solution Approach 1:
The patent performs preliminary filtering and feasibility assessment during the proposal generation process itself, rather than as a separate post-processing step. This allows the model to quickly eliminate infeasible options early in the generation process, maintaining high productivity while ensuring that only feasible proposals are presented to users.
Data Source
Figure 1~2
Figure 3~4
Figure 5(a)~6(b)
AI summary
A training program of a machine learning model outputs a proposal for obtaining a desired result, the training program of a machine learning model causes a computer to execute a process including: acquiring training data including a plurality of attributes; acquiring constraint condition data of the attributes; calculating first information regarding prediction accuracy of the machine learning model based on the training data; calculating second information regarding feasibility of the proposal based on the training data and the constraint condition data; calculating an evaluation index based on the first information and the second information; and training the machine learning model based on the evaluation index.