ML Campaign Segmentation De-Biasing Across Workflow Stages
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine-learning models often produce biased results due to skewed training data sets, sub-optimal model architectures, or bias-inducing post-processing, which can lead to legal and reputational risks, especially when influencing interactions with customers.
Innovation Solution
A system and method for detecting and mitigating bias in computational workflows by analyzing multiple class attributes, applying bias metrics, and using de-biasing techniques at various stages of the workflow, including pre-processing, in-processing, and post-processing, while balancing fairness with accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If larger training data sets are used to improve model accuracy, then model accuracy is improved, but the complexity of interpreting the model increases
Solution Approach 1:
The patent segments the analysis of training data by class attributes, examining each attribute separately to identify bias. This allows the system to handle large datasets while maintaining interpretability through structured analysis of individual attribute contributions to model predictions.
Solution Approach 2:
The patent introduces bias metrics as intermediary measurements that mediate between the complex model outputs and human interpreters. These metrics provide a bridge, translating complex model behavior into measurable bias indicators that can be monitored and managed without requiring direct interpretation of the entire model.
2Measurement precision
If more complex machine-learning models are developed to improve result accuracy, then result accuracy is improved, but the difficulty of detecting bias increases
Solution Approach 1:
The patent divides bias detection into segments by class attribute, allowing the system to analyze complex models attribute-by-attribute. This segmentation reduces the complexity of bias detection by breaking down the overall model into manageable components that can be individually assessed for bias.
Solution Approach 2:
The patent replaces manual, intuitive bias detection with automated machine learning-based bias metrics and detection algorithms. This substitution enables the system to handle complex models that would be difficult to analyze manually, using computational methods to automatically identify bias patterns.
3Reliability
If de-biasing techniques are applied to mitigate bias, then fairness is improved, but the complexity of the computational workflow increases
Solution Approach 1:
The patent segments the computational workflow into distinct stages: pre-processing (data cleaning and transformation), in-processing (model training with bias awareness), and post-processing (output adjustment). This segmentation allows de-biasing techniques to be applied systematically at each stage, managing complexity through structured implementation.
Solution Approach 2:
The patent implements dynamic de-biasing techniques that can adapt to different models and data characteristics. The system can adjust the degree and type of de-biasing applied based on the specific computational workflow, allowing flexibility that manages complexity by tailoring the approach to each particular case rather than using a fixed rigid process.
Data Source
AI summary
For at least a selected class attribute of the multiple class attributes, one or more bias metrics are determined that estimate a degree to which a particular workflow (having a set of processing stages) is biased in association with the class attribute. Each user of a set of users is associated with a set of user data to be processed by the particular workflow. At least one of the set of processing stages includes executing a machine-learning model. It can be detected that a bias-mitigation option corresponding to a specific class attribute has been selected. For each of at least two of the set of processing stages: a de-biasing technique is selected; and the processing stage is modified by applying the de-biasing technique. A modified version of the particular workflow (which includes the modified processing stages) is applied to each of a set of input data sets.