Machine Learning Post-Processing Bias Detection Using Historical Outputs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models suffer from biases due to prejudiced assumptions during development and data sampling, leading to skewed outputs and low prediction accuracy, which current systems fail to efficiently detect and mitigate across various stages of the model's lifecycle.
Innovation Solution
A system and method to detect and mitigate biases in machine learning models through a multi-stage framework, addressing pre-processing, in-processing, and post-processing stages by comparing datasets and model outputs with expected values, and applying targeted corrections to reduce biases and improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning models are trained using standard training datasets, then the models can be developed and deployed, but the models produce biased outputs and have low prediction accuracy due to prejudiced assumptions and biased data sampling
Solution Approach 1:
The system performs preliminary bias detection and mitigation actions at multiple stages: pre-processing (detecting biased data sampling and inappropriate features), in-processing (detecting biased assumptions during training), and post-processing (detecting biased outputs). By addressing biases before they fully manifest in model predictions, the system improves prediction accuracy while reducing harmful biased outputs.
Solution Approach 2:
The system implements continuous feedback loops where model outputs are evaluated for bias, and corrective actions are taken to mitigate detected biases. The system monitors prediction accuracy and bias levels, using this feedback to iteratively improve the model by adjusting data sampling, removing inappropriate features, and correcting biased assumptions throughout the model lifecycle.
2Reliability
If comprehensive bias detection and mitigation is performed across all stages of model development, then model accuracy and fairness improve, but computational resources and processing time increase
Solution Approach 1:
The system segments bias detection and mitigation into three distinct stages: pre-processing (data sampling and feature detection), in-processing (training assumption detection), and post-processing (output bias detection). Each stage has targeted detection mechanisms and corrective actions, allowing efficient resource allocation to specific bias types at appropriate development phases rather than applying uniform comprehensive processing throughout.
Solution Approach 2:
The system applies partial bias mitigation actions focused on the most critical bias types detected at each stage, rather than attempting to address all possible biases equally. The system prioritizes corrective actions that have the greatest impact on prediction accuracy and fairness, applying sufficient mitigation to address harmful biases while avoiding excessive processing of minor issues.
Data Source
AI summary
A system for mitigating biases during the testing of a machine learning model is disclosed. The system tests the machine learning model by inputting a set of real-world input data to the machine learning model, receiving a set of outputs from the machine learning model, and evaluating at least one of the set of outputs against a respective expected output. The respective expected output is determined based on a historical record associated with the set of real-world input data. The system determines that more than a threshold percentage of outputs from among the set of outputs differ from respective expected outputs. In response, the system determines that the machine learning model is biased. The system updates the machine learning model by updating one or more parameters of a neural network of the machine learning model.