Training Data Generation for Fairness-Aware Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for generating training data for fairness-aware machine learning models struggle to accurately identify and correct for discriminatory attributes, leading to potential biases in determination results due to the use of uniform threshold values that do not consider contextual dependencies.
Innovation Solution
A training data generation device that calculates a contribution level for each attribute, determines a reference value based on contextual considerations, and adjusts attribute values to reduce bias, using a process involving evaluation by multiple evaluators to establish agreed discrimination levels and modify data labels accordingly.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If uniform threshold values are used to identify discriminatory attributes, then the detection process is simple and fast, but the accuracy of discrimination detection is poor due to ignoring contextual dependencies
Solution Approach 1:
The patent segments the evaluation process into multiple stages: initial uniform threshold screening, followed by contextual dependency analysis for attributes near the threshold. This divides the problem into a simple filtering stage and a detailed analysis stage, improving accuracy without overwhelming complexity.
Solution Approach 2:
The patent applies different evaluation strategies to different attributes based on their characteristics. Attributes with clear discriminatory patterns use uniform thresholds, while attributes requiring contextual understanding receive specialized multi-evaluator assessment, optimizing both accuracy and efficiency.
2Measurement precision
If multiple evaluators are involved to establish agreed discrimination levels, then the accuracy of bias identification is improved, but the processing time and complexity increase
Solution Approach 1:
The patent performs preliminary uniform threshold screening before multi-evaluator contextual analysis. This preliminary action filters out clearly non-discriminatory attributes, so only attributes near the threshold require time-consuming multi-evaluator assessment, reducing overall processing time.
Solution Approach 2:
The patent applies the resource-intensive multi-evaluator process only partially—to attributes that fall within a specific range of the threshold—rather than to all attributes. This selective application maintains high accuracy where needed while minimizing time loss.
3Reliability
If attribute values are adjusted based on contextual dependencies, then fairness in determination results is improved, but the complexity of data processing increases
Solution Approach 1:
The patent applies contextual adjustment only to attributes identified as having discriminatory potential through initial screening. Most attributes undergo simple threshold comparison, while only specific attributes require complex contextual analysis and adjustment, maintaining overall system efficiency.
Solution Approach 2:
The patent changes the evaluation parameters dynamically: using fixed uniform thresholds for initial screening, then switching to flexible contextual parameters for attributes requiring detailed analysis. This parameter adaptation resolves the contradiction between simplicity and accuracy.
Data Source
AI summary
A non-transitory computer-readable storage medium storing a training data generation program for causing a computer to perform processing including: receiving an evaluation value for a value calculated on a basis of a number of data for each attribute included in a plurality of data; determining a reference value for each attribute on a basis of the received evaluation value and the number of data for each attribute; and generating training data for machine learning by changing the attribute of at least partial data of the plurality of data according to the reference value for each attribute.