Machine Learning Model Training to Exclude Undesired Factors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models often incorporate undesired factors such as age, sex, and race in their analyses due to biased training data, leading to unfair outcomes in insurance underwriting.
Innovation Solution
A method to train machine learning models to identify and exclude undesired factors by combining a neural network with a linear model or training the neural network to recognize and ignore these factors, ensuring fair analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If machine learning models are trained with unstructured data including all available information, then the model can consider comprehensive data for correlations and predictions, but the model may incorporate undesired factors such as age, sex, ethnicity, and race leading to bias and discrimination
Solution Approach 1:
The patent extracts and removes undesired factors (protected characteristics such as age, sex, ethnicity, race) from the training data and model considerations. This is achieved by identifying these factors and explicitly excluding them from the features the machine learning model uses for predictions, thereby eliminating the source of bias while preserving useful information.
Solution Approach 2:
The patent segments the training data into desired features and undesired factors (protected characteristics). By separating these components, the model can process comprehensive data while systematically excluding the harmful segments that lead to discrimination, allowing selective feature engineering without losing overall data comprehensiveness.
2Measurement precision
If machine learning models are trained to consider all available data including background information, then the model can make more accurate predictions, but the model may learn illegitimate correlations with protected characteristics
Solution Approach 1:
The patent converts the potentially harmful effect of comprehensive data (which may contain biased correlations) into a benefit by using the full dataset for training while simultaneously applying constraints that prevent the model from learning illegitimate correlations. The comprehensive data provides rich patterns for accurate prediction, while the exclusion of protected characteristics ensures fairness.
3Object-affected harmful factors
If the machine learning model is constrained to use only structured data, then undesired factors are limited, but the model loses access to valuable unstructured information that could improve analysis
Solution Approach 1:
The patent applies local quality by treating different types of data differently: structured data is processed with standard feature engineering, while unstructured data (images, text, audio) is processed through specialized pipelines that extract relevant features while explicitly excluding protected characteristics. This allows the model to leverage the unique value of each data type without uniformly applying restrictions that would lose information.
Data Source
AI summary
A method of training and using a machine learning model that controls for consideration of undesired factors which might otherwise be considered by the trained model during its subsequent analysis of new data. For example, the model may be a neural network trained on a set of training images to evaluate an insurance applicant based upon an image or audio data of the insurance applicant as part of an underwriting process to determine an appropriate life or health insurance premium. The model is trained to probabilistically correlate an aspect of the applicant's appearance with a personal and/or health-related characteristic. Any undesired factors, such as age, sex, ethnicity, and/or race, are identified for exclusion. The trained model receives the image (e.g., a “selfie”) of the insurance applicant, analyzes the image without considering the identified undesired factors, and suggests the appropriate insurance premium based only on the remaining desired factors.