Dynamic Outlier Bias Reduction in Machine Learning Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models often suffer from bias due to flawed sample/training data selection, which can lead to inaccurate predictions and determinations.
Innovation Solution
The implementation of dynamic outlier bias reduction methods, which involve receiving a training data set, applying a machine learning model to determine predicted values, generating error sets, and utilizing a data selection vector to identify non-outlier data elements, thereby updating model parameters and iteratively refining predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If training data includes all available data elements, then the model has more data to learn from, but bias from outliers degrades prediction accuracy
Solution Approach 1:
The patent extracts and removes outlier data elements from the training dataset that cause bias. By identifying data points with excessive influence on model parameters and removing them, the system maintains the benefits of large training data volume while eliminating the harmful bias that degrades prediction accuracy.
Solution Approach 2:
The patent changes the parameter selection criteria for training data by introducing influence measures and bias criteria. Instead of using all data equally, the system dynamically adjusts which data points are included based on their statistical influence and bias characteristics, transforming the training data selection process to optimize both data utilization and prediction accuracy.
2Productivity
If the model iteratively refines parameters using all training data, then convergence may be faster, but bias propagation maintains or worsens prediction errors
Solution Approach 1:
The patent introduces dynamic data selection that changes during the iterative training process. The training dataset is dynamically adjusted based on influence measures calculated at each iteration, allowing the model to adaptively focus on high-quality data points while maintaining efficient convergence through iterative refinement.
Solution Approach 2:
The system implements feedback mechanisms where prediction errors and influence measures from previous iterations inform the selection of training data for subsequent iterations. This feedback loop allows the model to progressively eliminate bias sources while maintaining training efficiency through targeted data selection.
3Measurement precision
If outliers are removed from training data, then prediction accuracy improves, but the complexity of data processing and model training increases
Solution Approach 1:
The patent replaces complex manual or heuristic outlier detection methods with automated statistical measures and influence functions. By using mathematical formulations to automatically identify and remove outliers based on their influence on model parameters, the system reduces processing complexity while maintaining high prediction accuracy.
Data Source
AI summary
Systems and methods include processors for receiving training data for a user activity; receiving bias criteria; determining a set of model parameters for a machine learning model including: (1) applying the machine learning model to the training data; (2) generating model prediction errors; (3) generating a data selection vector to identify non-outlier target variables based on the model prediction errors; (4) utilizing the data selection vector to generate a non-outlier data set; (5) determining updated model parameters based on the non-outlier data set; and (6) repeating steps (1)-(5) until a censoring performance termination criterion is satisfied; training classifier model parameters for an outlier classifier machine learning model; applying the outlier classifier machine learning model to activity-related data to determine non-outlier activity-related data; and applying the machine learning model to the non-outlier activity-related data to predict future activity-related attributes for the user activity.