Machine Learning Model Bias Correction via Proxy Feature Retraining
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models often exhibit bias due to systematic errors in training data or processes, which can lead to inaccurate results, and existing methods lack efficiency in identifying and correcting these biases, especially in complex data sets.
Innovation Solution
A computer-implemented method generates a candidate machine learning model, tests it for bias, and if biased, generates a proxy feature, calculates weights, updates the data set, and re-trains the model to eliminate bias, using performance metrics and impact ratios to ensure accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If machine learning models are trained using complex training data sets, then the models can handle diverse and large-scale data, but systematic errors and bias in the training process may result in flawed models that produce inaccurate results
Solution Approach 1:
The patent implements a feedback mechanism where the trained machine learning model is tested to generate performance metrics indicating bias values. These metrics feed back into the training process by generating proxy features that are added to the model and used to update training data weights, creating a closed-loop system that continuously reduces bias while maintaining the model's ability to handle diverse data
Solution Approach 2:
The patent changes the parameters of the training process by dynamically adjusting data weights based on bias metrics. Specifically, it calculates weights for training data samples based on the bias values identified during testing, and retrained the model with these adjusted weights, thereby modifying the training parameters to eliminate systematic errors while preserving adaptability
2Device complexity
If traditional methods are used to identify bias in machine learning models, then the process may be simple to implement, but it is inefficient and difficult to identify errors in complex training data sets
Solution Approach 1:
The patent implements a self-service mechanism where the machine learning model evaluates itself for bias through automated testing that generates performance metrics. The system automatically identifies bias without external intervention, generates proxy features based on the bias values, and retrains itself with updated weights, making the bias identification and correction process autonomous and highly efficient for complex data sets
3Reliability
If machine learning models are retrained multiple times to eliminate bias, then the model accuracy and fairness improve, but the time and computational resources required increase
Solution Approach 1:
The patent applies preliminary action by generating proxy features based on bias values before the actual retraining process. These proxy features are added to the model in advance and used to calculate data weights that guide the subsequent retraining, preparing the model proactively to reduce bias more efficiently and reduce the time required for iterative corrections
Data Source
AI summary
A computer-implemented method includes generating a candidate machine learning model. The candidate machine learning model is configured to generate probability scores for members. The method includes training the candidate machine learning model using an initial data set, testing the candidate machine learning model to generate performance metrics indicative of a bias value correlated to the candidate machine learning model, and determining whether the performance metrics are below a threshold. In response to determining that the performance metrics are below the threshold the method includes generating a proxy feature based on the bias value, adding the proxy feature to the candidate machine learning model, calculating weights based on the bias value, updating the initial data set using the weights to generate an updated data set, and retraining the candidate machine learning model using the updated data set.