Machine Learning Model Collection for Explainable Predictions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models, especially in safety-critical domains, lack the ability to provide alternative perspectives and justified explanations for their predictions, particularly in 'corner cases' where the model is not confident.
Innovation Solution
A method utilizing a collection of machine learning models with competing strategies to yield multiple predictions and explanations for a dataset, allowing for contrastive explanations and model voting with explainability features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single machine learning model is used to achieve high performance, then prediction accuracy is improved, but explainability and alternative perspectives are lost
Solution Approach 1:
The patent divides a single model into multiple specialized models, each trained with different strategies (e.g., different loss functions, data subsets, or architectural choices). This segmentation allows each model to excel at specific aspects of the prediction task while collectively providing diverse perspectives and explanations for the overall system.
2Device complexity
If a single machine learning model is used to simplify the system, then device complexity is reduced, but the ability to provide justified explanations in corner cases deteriorates
Solution Approach 1:
The system dynamically selects and weights different models based on the input characteristics and confidence levels. In corner cases where models are uncertain, the system can activate alternative models or provide multiple explanations, making the system adapt its complexity based on the situation rather than being statically simple or complex.
3Loss of information
If multiple machine learning models are used to provide alternative perspectives, then explanation quality is improved, but device complexity increases
Solution Approach 1:
The patent merges multiple specialized models into a unified model collection framework that manages them collectively. This framework provides standardized interfaces for prediction, explanation generation, and model selection, allowing the system to leverage multiple models while presenting a unified, manageable interface to users and reducing operational complexity.
4Adaptability or versatility
If machine learning models are continuously updated to maintain performance, then adaptability is improved, but the need for justified results in safety-critical domains worsens due to black-box nature
Solution Approach 1:
The system incorporates feedback loops where model predictions and explanations are continuously evaluated. When performance degradation is detected or in safety-critical scenarios, the system can trigger retraining, model switching, or enhanced explanation generation. This feedback mechanism ensures continuous adaptability while maintaining accountability through explainability.
Data Source
AI summary
A method for explanation of machine learning results based on using a model collection includes training at least two machine learning models with at least two competing strategies for the at least one dataset; and using the least two machine learning models to yield at least two different predictions and/or at least two explanations for the at least one dataset.