Teacher Data Editing Assistance for AI Fairness
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for improving fairness in machine learning models by increasing the number of teacher data points do not always reduce sensitive determinations, and are limited to binary classification problems, failing to address regression problems and effectively mitigate discrimination in AI models.
Innovation Solution
A teacher data editing support system that calculates the contribution of sensitive attributions to correct answers, visually presents evaluation information, and allows for designated changes to the correct answers in teacher data to reduce sensitive determinations, applicable to both regression and classification problems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the number of teacher data points is increased to improve prediction accuracy and fairness, then the model's prediction accuracy improves, but the sensitive determination may not be reduced and discrimination may persist
Solution Approach 1:
The system extracts and isolates the sensitive attribution component from the teacher data by calculating its specific contribution to correct answers. This allows the sensitive information to be separately quantified and adjusted without discarding the entire data point, thereby reducing sensitive determination while preserving prediction accuracy.
Solution Approach 2:
The system changes the parameter of the correct answer by adjusting it based on the calculated contribution of sensitive attribution. This parameter adjustment directly reduces the influence of sensitive information on the model's predictions while maintaining the overall quality and accuracy of the teacher data.
2Adaptability or versatility
If data preprocessing techniques are applied to rewrite correct answers for improving fairness index, then fairness is improved for binary classification, but the method cannot be applied to regression problems and other problem types
Solution Approach 1:
The system achieves universality by designing a contribution calculation mechanism that works across different problem types including both classification and regression. The core functionality of calculating sensitive attribution contribution and adjusting correct answers is applicable universally, making the system versatile rather than limited to binary classification only.
Solution Approach 2:
The system applies parameter changes to the correct answer based on the calculated contribution of sensitive attribution. This approach is universally applicable to different problem types because it operates on the fundamental relationship between sensitive attributes and prediction outcomes, regardless of whether the problem is classification or regression.
3Quantity of substance
If perturbation images are generated and added to teacher data to increase data quantity, then the number of teacher data points increases, but this does not reduce the influence of sensitive attribution if the original image is affected by sensitive attribution
Solution Approach 1:
Instead of merely increasing data quantity through perturbation, the system extracts and quantifies the specific contribution of sensitive attribution to correct answers. This extraction allows for targeted adjustment of the sensitive information's influence while preserving the utility of the teacher data points.
Solution Approach 2:
The system changes the parameter of the correct answer by adjusting it according to the calculated contribution of sensitive attribution. This parameter change directly reduces the influence of sensitive information regardless of the number of data points, making the quantity of data less relevant to solving the discrimination problem.
Data Source
AI summary
A teacher data editing support system includes a determination unit that receives teacher data including sensitive attribution as a variable that potentially causes discrimination, a feature as a variable to be used for prediction, and a correct answer, and calculates contribution as an index indicating contribution of the sensitive attribution to the correct answer, a display unit that visually presents evaluation information indicating a relationship between a level of changing the correct answer in the teacher data and a level of deviation of the correct answer from an initial value or a discrimination level based on the contribution, and an editing unit that accepts designation of how much the correct answer is changed, changes the correct answer in the teacher data in response to the designation, and outputs the changed teacher data.