Combined Linear Surrogate Model for Consistent ML Interpretation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning interpretability techniques struggle with consistency when approximating complex classifier boundaries with linear surrogate models, leading to varying interpretations across multiple runs.
Innovation Solution
The method combines two or more linear models into a combined linear model by selecting concurrent linear models and an observation of interest, determining a neighborhood of feature vectors, and generating a combined model as either a mean local linear surrogate or a linearly weighted local surrogate model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single linear surrogate model is used to approximate complex classifier boundaries, then the model complexity is kept simple, but the interpretation consistency deteriorates across multiple runs
Solution Approach 1:
The patent combines multiple linear surrogate models into an ensemble to improve interpretation consistency. Specifically, it generates multiple linear models from different sampling schemes and combines their predictions through averaging or voting mechanisms, thereby maintaining simplicity while achieving reliable and consistent interpretations across multiple runs.
2Reliability
If multiple linear models are combined into an ensemble, then the interpretation consistency is improved, but the device complexity increases
Solution Approach 1:
The patent segments the model generation process into distinct phases: generating multiple linear models from different sampling schemes, evaluating their individual performances, and then combining them through weighted or unweighted averaging. This segmentation allows for systematic management of the ensemble complexity while maintaining interpretation consistency.
Solution Approach 2:
The patent employs parameter changes by varying sampling schemes and model generation parameters to create diversity among the linear models in the ensemble. By changing parameters such as sampling density, neighborhood size, and model training configurations, the system generates varied models that when combined, provide consistent and robust interpretations.
3Measurement precision
If a sampling scheme is used to create linear boundaries, then the local approximation accuracy is improved, but the boundary variability worsens across multiple runs
Solution Approach 1:
The patent applies beforehand cushioning by generating multiple linear models in advance using different sampling schemes before making final predictions. This pre-generation of diverse models cushions against the variability that would otherwise occur in single-model approaches, ensuring stable and consistent boundaries across multiple runs while maintaining local approximation accuracy.
Data Source
AI summary
Systems and methods are described to combine two or more linear models into a combined linear model. Two or more linear models and an observation of interest are selected. The linear models are concurrent with respect to the observation of interest. The observation of interest includes a class value and a feature vector. A distance is selected, and a plurality of feature vectors are selected that are within the distance of the feature vector associated with the observation of interest. These feature vectors are input to the selected linear models and a plurality of class values are generated. These class values and the selected feature vectors are used to generate a combined linear model. The combined model is either a mean local linear surrogate model or a linearly weighted local surrogate model.