Machine Learning Bias Correction via Input Substitution and Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning systems fail to guarantee fairness in prediction models, leading to unintended biases that affect decisions, such as loan approvals, as they do not consider biases in learned models during operation.
Innovation Solution
A system that includes a substitution unit to rewrite input values based on predetermined conditions, a preservation unit to store prediction results, a reception unit for feedback, and a generation unit to relearn the model using feedback data, ensuring fairness by generating a new model that corrects biases.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a learned model is used for prediction without bias correction, then prediction speed is maintained, but fairness and accuracy deteriorate due to unintended biases
Solution Approach 1:
The system performs preliminary actions by substituting input values before prediction to prevent bias from affecting the prediction result. The substitution unit replaces potentially biased input values with alternative values that reflect fairer assessment criteria, ensuring fairness is addressed before the prediction is made rather than attempting to correct biases afterward.
Solution Approach 2:
The system introduces an intermediary mechanism (substitution unit) that mediates between the original input data and the learned model. This intermediary layer transforms biased inputs into fairer representations without requiring modification of the underlying learned model, thus improving fairness while maintaining relatively simple system architecture.
2Reliability
If input values are substituted to correct bias, then fairness is improved, but processing time increases due to additional substitution steps
Solution Approach 1:
The system applies partial action by selectively substituting only those input values that are identified as potentially biased, rather than transforming all input data. This targeted approach corrects fairness issues in critical areas while minimizing the overall processing overhead associated with comprehensive data transformation.
3Measurement precision
If the learned model is retrained with feedback data, then prediction accuracy improves, but computational cost and training time increase
Solution Approach 1:
The system implements a feedback mechanism where prediction results are collected and used to generate feedback data that informs subsequent model retraining. This closed-loop feedback allows the model to continuously improve accuracy by learning from actual outcomes, ensuring that computational resources are invested in training only when and how it will improve prediction performance.
Solution Approach 2:
Before full model retraining is initiated, the system performs preliminary actions by substituting biased input values in the feedback data. This preliminary correction ensures that the training data itself is fair and accurate, preventing the model from learning biased patterns in the first place, thereby reducing the need for extensive retraining and associated computational costs.
Data Source
AI summary
A system managing a first model generated by machine learning, the system comprising: at least one processor and at least one memory functioning as: a substitution unit configured to substitute a first input value included in an input with a second input value when a first prediction result which is a prediction result obtained using the input including the first input value by the first model satisfies a predetermined condition; a preservation unit configured to preserve a second prediction result which is a prediction result obtained using an input obtained through the substitution by the first model; a reception unit configured to receive feedback on the second prediction result; and a generation unit configured to generate a second model by performing machine learning using first learning data formed by the input including the first input value and the second prediction result based on the received feedback.