Tree-Based Ruleset Generation for Black-Box ML Explainability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Complex deep learning models are challenging to interpret due to their complexity, leading to a need for improved techniques in machine learning explainability (MLX) that can generate understandable and efficient explanations for model predictions.
Innovation Solution
Techniques for generating an improved initial set of explanatory rules from a decision tree are developed, which are then optimized to provide a more accurate and interpretable ruleset for MLX, using tree-based representations and information gain to select influential features and split values, resulting in improved fidelity and reduced rule count.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If complex deep learning models are used to solve complex problems, then model capacity and performance are improved, but model interpretability and explainability deteriorate
Solution Approach 1:
The patent segments the complex deep learning model into multiple decision stumps, each handling a specific feature condition. This segmentation transforms the monolithic complex model into a collection of simple, interpretable decision rules that collectively approximate the model's behavior, resolving the contradiction between model capacity and interpretability.
Solution Approach 2:
The patent introduces an intermediary process that extracts feature importances from the complex model and uses these to generate decision stumps. This intermediary layer bridges the gap between the black-box complex model and the interpretable ruleset, allowing the system to maintain high performance while providing explainability.
2Measurement precision
If more rules are generated to improve explanation accuracy, then coverage and fidelity are improved, but computation time and rule count increase
Solution Approach 1:
The patent performs preliminary actions by pre-computing feature importances and generating decision stumps before the actual explanation task. This preprocessing step creates a ready-to-use ruleset that can quickly explain model predictions without requiring extensive computation during the explanation phase, thus improving accuracy while reducing time loss.
Solution Approach 2:
The patent generates a large number of candidate decision stumps initially (excessive action) and then selects the most important ones through optimization. This approach ensures high coverage and fidelity by considering all possible rules, while the subsequent selection process reduces the final rule count to manageable levels, balancing accuracy and computation time.
Data Source
AI summary
Herein are techniques to generate candidate rulesets for machine learning (ML) explainability (MLX) for black-box ML models. In an embodiment, an ML model generates classifications that each associates a distinct example with a label. A decision tree that, based on the classifications, contains tree nodes is received or generated. Each node contains label(s), a condition that identifies a feature of examples, and a split value for the feature. When a node has child nodes, the feature and the split value that are identified by the condition of the node are set to maximize information gain of the child nodes. Candidate rules are generated by traversing the tree. Each rule is built from a combination of nodes in a tree traversal path. Each rule contains a condition of at least one node and is assigned to a rule level. Candidate rules are subsequently optimized into an optimal ruleset for actual use.