AI Bias Removal Method Search System
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing bias removal methods for AI models often compromise prediction accuracy and lack clarity on unconsidered improvement effects, with varying requirements for different business applications and multiple evaluation indexes.
Innovation Solution
A computer system that searches and presents a bias removal method for generating AI models, satisfying specific business requirements related to bias removal and prediction accuracy by managing a dataset, business requirement definition information, and bias removal method management information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a bias removal method is applied to an AI model, then the bias in the model is reduced, but the prediction accuracy of the model is reduced
Solution Approach 1:
The system changes parameters by applying different bias removal methods (such as rewriting objective variables, rewriting learning weights, converting features, or adding bias index terms to the Loss function) and their associated parameters to the AI model training process. By adjusting these parameters and methods, the system seeks to reduce bias while monitoring the impact on prediction accuracy, allowing for optimization of the trade-off between fairness and accuracy based on specific business requirements.
2Reliability
If a certain bias removal method is applied, then the bias is removed, but other improvement effects of the bias not considered in the bias removal method remain unknown
Solution Approach 1:
The system implements feedback by evaluating the AI model's performance across multiple dimensions after applying bias removal methods. It calculates various evaluation indexes including not only the primary bias removal metric but also other potential improvement effects. This feedback mechanism allows the system to identify and quantify previously unknown improvement effects, such as enhanced model generalization or reduced overfitting, that may result from applying bias removal techniques.
3Adaptability or versatility
If multiple bias removal methods are evaluated, then the comprehensive understanding of bias removal options is improved, but the complexity of the search process increases
Solution Approach 1:
The system segments the complex search process into distinct, manageable components: (1) selecting candidate bias removal methods from a predefined set, (2) applying each method to the training data, (3) evaluating performance using multiple evaluation indexes, and (4) selecting the optimal method based on business requirements. This segmentation reduces the overall complexity by breaking down the monolithic search problem into smaller, more tractable sub-problems that can be solved systematically.
Data Source
AI summary
A computer system receives a request including information for specifying a business requirement, executes search processing for searching for a bias removal method used to generate an AI model that satisfies the specified business requirement, and generates a report including information on the bias removal method used to generate the AI model that satisfies the specified business requirement. In the search processing, processing of selecting the bias removal method and processing of executing learning processing of generating the AI model by using the selected bias removal method are repeatedly executed until the specified business requirement is satisfied.