Counterfactual Explainer Using Pre-computed Interpolations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current counterfactual explainers face challenges such as limited generalization, high computational cost, limited interpretability, and difficulty in handling sparse features, often failing to generate accurate and diverse explanations for machine learning models.
Innovation Solution
The AutoMLx Counterfactual Explainer (ACE) employs k-d trees for efficient nearest neighbor search, uses batch processing, and combines discrete and continuous interpolation techniques to generate diverse and sparse counterfactuals, ensuring valid explanations are found for every test instance with minimal feature value changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If counterfactual explainers use traditional methods (surrogate models, gradient processing, decision trees), then they can generate explanations, but they suffer from limited generalization, high computational cost, and poor scalability
Solution Approach 1:
The patent pre-computes and stores counterfactual explanations for training data instances during the training phase. This preliminary action allows the system to retrieve and reuse pre-computed explanations for test instances, eliminating the need for expensive real-time counterfactual search and significantly improving computational efficiency while maintaining explanation accuracy
Solution Approach 2:
The patent creates a library of pre-computed counterfactual explanations that can be copied and reused for different test instances. By storing explanations in a database and retrieving appropriate explanations based on input similarity, the system avoids redundant computation and achieves high scalability without sacrificing explanation quality
2Measurement precision
If counterfactual explainers perform case-by-case search, then they can generate explanations for specific instances, but time complexity grows linearly for sets of inferences
Solution Approach 1:
The system performs counterfactual search and explanation generation during the training phase on training data, storing the results in a database. During inference, the system simply queries this pre-computed database, reducing processing time from linear growth to constant time per instance while maintaining explanation validity through the pre-computed nature of the explanations
Solution Approach 2:
The patent introduces a database as an intermediary between the input data and the explanations. This database stores pre-computed counterfactual explanations and enables fast retrieval based on input features, acting as a mediator that eliminates the need for real-time computation and achieves O(1) query time instead of linear time complexity
3Adaptability or versatility
If explainers generate diverse explanations, then interpretability improves, but computational resources and time increase
Solution Approach 1:
The patent pre-computes multiple diverse counterfactual explanations for each training instance and stores them in the database. During inference, the system retrieves these pre-generated diverse explanations without performing additional computation to generate diversity, thus maintaining explanation variety while minimizing real-time computational resource consumption
Solution Approach 2:
The system copies pre-computed diverse explanations from the database rather than generating them anew for each query. This copying approach preserves explanation diversity and versatility while eliminating the computational burden of regenerating diverse explanations, significantly reducing energy and resource consumption during inference
Data Source
AI summary
A computer stores a reference corpus that consists of many reference points that each has a respective class. Later, an expected class and a subject point (i.e. instance to explain) that does not have the expected class are received. Multiple reference points that have the expected class are selected as starting points. Based on the subject point and the starting points, multiple discrete interpolated points are generated that have the expected class. Based on the subject point and the discrete interpolated points, multiple continuous interpolated points are generated that have the expected class. A counterfactual explanation of why the subject point does not have the expected class is directly generated based on continuous interpolated point(s) and, thus, indirectly generated based on the discrete interpolated points. For acceleration, neither way of interpolation (i.e. counterfactual generation) is iterative. Generated interpolated points can be reused to amortize resources consumed while generating counterfactuals.