Feature Relevance Visualization for AI Explainability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current artificial intelligence systems, particularly in decision-making processes like misappropriation detection, lack explainability and interpretability, leading to challenges in reliability and compliance with regulatory requirements, as the accuracy of these systems is inversely proportional to their explainability and interpretability.
Innovation Solution
A system that optimizes feature relevance visualization through logical and machine learning-based grouping, combined with analyst input, and employs neural network output correction using layer-wise relevance propagation and batch normalization to enhance explainability, providing user-friendly visualizations and correcting side effects in neural network outputs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If neural network models are used to improve detection accuracy in misappropriation detection systems, then accuracy is improved, but explainability and interpretability deteriorate
Solution Approach 1:
The patent introduces feature relevance visualization as an intermediary component between the neural network model and the user. This visualization technique translates the internal workings of the complex neural network into an interpretable format that shows which features most influenced the detection decision, thereby maintaining detection accuracy while recovering explainability information
Solution Approach 2:
The patent changes the parameter representation from raw neural network internal states to feature relevance scores and visualizations. By transforming the output parameters of the neural network into human-understandable feature importance metrics, the system maintains the accuracy benefits of neural networks while providing explainable results
2Loss of information
If feature relevance visualization is enhanced to improve explainability, then interpretability is improved, but system complexity increases
Solution Approach 1:
The patent extracts only the essential explainability information (feature relevance scores and top contributing features) from the complex neural network processing, rather than attempting to visualize all internal states. This selective extraction provides interpretability while avoiding the complexity of comprehensive model visualization
Solution Approach 2:
The patent applies different levels of detail and visualization techniques to different parts of the feature space based on their relevance. High-relevance features receive more detailed visualization and analysis, while low-relevance features are summarized or omitted, creating an explainability system with varying local quality that reduces overall complexity
Data Source
AI summary
A system for feature relevance visualization optimization is provided. The system comprises a controller configured for modifying placement of features in a relevance visualization. The controller is further configured to: receive interaction data comprising one or more features positioned in the relevance visualization, wherein the one or more features are defined and measurable properties of the interaction data; construct a logical grouping of the one or more features based on a type of each of the one or more features, wherein similar features are collocated in the relevance visualization; construct a machine learning-based grouping of the one or more features based on relevance calculations for the one or more features; combine the logical grouping and the machine learning-based grouping to generate a combined feature placement, wherein the one or more features are repositioned in the relevance visualization; and output the relevance visualization having the combined feature placement.