Version-Based Model Explainability Through Result Comparison
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current model output explanations do not provide insights into how different stages of model development or training data sets affect the model's results, limiting user understanding and trust in model functionality.
Innovation Solution
A method that executes a first version of a model, selects a second version based on input data attributes, and constructs a natural language explanation of the difference between the two execution results using a natural language processing engine.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If model versions are updated with additional training data or attributes, then model accuracy and functionality are improved, but user understanding of model behavior deteriorates due to lack of explanation about version differences
Solution Approach 1:
The system segments the model evolution into distinct versions, each with specific training data and attributes. By comparing individual version differences rather than treating the model as a monolithic entity, the system can provide targeted explanations about what changed between versions, maintaining user understanding while allowing continuous model improvement.
Solution Approach 2:
The system introduces an intermediary explanation layer that mediates between the complex model version differences and user understanding. This intermediary generates natural language explanations that translate technical model changes into comprehensible information about how and why predictions differ between versions, preserving both model accuracy improvements and user comprehension.
2Loss of information
If multiple model versions are executed to provide explanations, then explainability is improved, but computational resources and time are consumed
Solution Approach 1:
The system performs preliminary actions by pre-identifying and caching model versions that are likely to be relevant for comparison based on input data attributes. Rather than executing all possible version comparisons, the system proactively selects a subset of versions that will provide meaningful explanations, reducing computational overhead while maintaining explainability quality.
Solution Approach 2:
The system dynamically adjusts comparison parameters such as the threshold for what constitutes a significant prediction difference and the number of versions to compare. By changing these parameters based on the specific input and context, the system optimizes the balance between providing sufficient explainability and minimizing computational time and resources.
Data Source
AI summary
A first version of a model specified by a model execution request is executed, producing a first execution result. A second version of the model is selected according to an input data attribute specified by the model execution request. The second version of the model is executed, producing a first execution result. Using a natural language processing engine, responsive to the first execution result and the second execution result differing by more than a threshold amount, a natural language explanation of a difference between the first execution result and the second execution result is constructed.