Constrained Decision Trees for Bias-Resistant ML Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models face challenges in generating accurate predictions when trained on unlabeled, few-labeled, or unreliable data, often due to subjective human-generated labels and biased feature importance.
Innovation Solution
The use of constrained decision trees and judgmental sampling techniques, such as multivariate stratified sampling and feature ranking, to select observations and prioritize features based on expert opinions, allowing for the generation of machine learning models that reduce subjectivity and bias.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If human experts generate labels for machine learning training data, then the model can learn from labeled data, but the process consumes substantial time and resources and introduces subjectivity and bias
Solution Approach 1:
The patent uses synthetic data generation to create artificial labeled examples that mimic real data distributions. Instead of manually labeling each data point, the system generates synthetic labeled data that captures the underlying patterns, thereby reducing time consumption while maintaining label quality for training the machine learning model.
Solution Approach 2:
The system employs automated algorithms to generate labels and features without human intervention. The machine learning model itself participates in the labeling process through self-training mechanisms, where it generates its own training data and labels, eliminating the need for external human experts and reducing both time and subjectivity.
2Measurement precision
If human experts generate labels for machine learning training data, then the model can learn from labeled data, but the process is expensive due to substantial time and resource consumption
Solution Approach 1:
The patent uses synthetic data generation to create artificial labeled examples that mimic real data distributions. Instead of manually labeling each data point, the system generates synthetic labeled data that captures the underlying patterns, thereby reducing time consumption while maintaining label quality for training the machine learning model.
Solution Approach 2:
The system employs automated algorithms to generate labels and features without human intervention. The machine learning model itself participates in the labeling process through self-training mechanisms, where it generates its own training data and labels, eliminating the need for external human experts and reducing both time and subjectivity.
3Measurement precision
If human experts generate labels, then the model receives labeled data for supervised learning, but the labels reflect biases of the humans who generate them
Solution Approach 1:
The system employs automated algorithms to generate labels and features without human intervention. The machine learning model itself participates in the labeling process through self-training mechanisms, where it generates its own training data and labels, eliminating the need for external human experts and reducing both time and subjectivity.
Solution Approach 2:
The patent introduces synthetic data as an intermediary between real data and model training. Instead of directly using human-generated labels, the system creates synthetic labeled data that serves as a mediator, filtering out human biases while preserving the essential patterns and relationships needed for effective model training.
4Measurement precision
If features are selected based on human expert opinion, then the model can incorporate domain knowledge, but different experts may have different opinions on feature importance
Solution Approach 1:
The patent implements dynamic feature selection where the importance of features is not fixed but adapts based on the data and model performance. The system automatically adjusts feature weights and selections during training, allowing the model to discover the most relevant features empirically rather than relying on static expert opinions, thereby reducing complexity from multiple expert perspectives.
Solution Approach 2:
The patent introduces synthetic data as an intermediary between real data and model training. Instead of directly using human-generated labels, the system creates synthetic labeled data that serves as a mediator, filtering out human biases while preserving the essential patterns and relationships needed for effective model training.
Data Source
AI summary
Techniques are described herein that are capable of generating a machine learning model based on constrained decision trees using a judgmental sample and feature ranking. A judgmental sample including observations, which include respective subsets of features, is generated. The observations are selected using multivariate stratified sampling. Important subsets of the features are determined based on each important subset being designated as more important than the other features by a respective individual. A score is determined for each feature, indicating a proportion of the important subsets that includes the respective feature. A highest scored feature is identified. Constrained decision trees having respective first splits are generated, based on respective subsets of the observations. A proportion of the first splits corresponding to the highest scored feature is based at least on the score of the highest scored feature. A machine learning model is generated based at least on the constrained decision trees.