Automated XAI Relevance Scoring for Machine Decision Explainability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing XAI solutions are limited by the type of data processed and lack automation in explainability processes, particularly for structured, semi-structured, and unstructured text data, which is crucial in critical domains like medical, pharma, and finance where small errors can have significant impacts.
Innovation Solution
A method and system utilizing a deep neural network (DNN) model with automated XAI prediction processes to generate explainable models (MXAI) through layerwise relevance propagation, computing feature contributions, and displaying relevance scores on a user interface, supporting structured, semi-structured, and unstructured data types.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated XAI prediction process is implemented, then explainability process is automated and productivity is improved, but device complexity increases due to deep neural network training and multiple iterations
Solution Approach 1:
The system performs self-service by automatically training deep neural networks, executing multiple iterations of predictions, and generating explainability metrics without requiring manual intervention. The automated XAI prediction process handles the entire workflow from data input to explanation generation, eliminating the need for manual model training and analysis procedures.
Solution Approach 2:
The system manages complexity by dynamically adjusting training parameters such as the number of iterations, learning rates, and network architecture based on the specific task requirements and data characteristics. This allows the system to adapt the complexity level to match the problem domain while maintaining automation.
2Measurement precision
If deep neural network model is trained through multiple iterations, then prediction accuracy and relevance score computation are improved, but computational resources and time consumption increase
Solution Approach 1:
The system implements feedback mechanisms where predictions from one iteration are used to refine the model for subsequent iterations. The relevance scores computed in each iteration provide feedback on feature importance, allowing the system to converge on accurate predictions and stable explainability metrics more efficiently.
Solution Approach 2:
The system executes multiple iterations of training and prediction, performing more computational steps than strictly minimum required. This excessive action ensures thorough convergence of the neural network and stability of relevance scores, particularly important for critical domains where high accuracy is paramount.
3Loss of information
If layerwise relevance propagation is used to compute feature contributions, then explainability and feature contribution analysis are improved, but processing complexity and computational overhead increase
Solution Approach 1:
The system segments the neural network into distinct layers and computes relevance scores for each layer separately through layerwise relevance propagation. This segmentation allows the complex task of feature contribution analysis to be broken down into manageable steps, making the computation more efficient and the results more interpretable.
Solution Approach 2:
The relevance score computation acts as an intermediary mechanism between the neural network's internal representations and the final explainability output. This intermediary process translates complex neural computations into human-interpretable feature importance metrics without requiring direct inspection of the network's internal state.
4Adaptability or versatility
If automated XAI process handles multiple data types, then versatility and adaptability are improved, but system complexity and difficulty of processing increase
Solution Approach 1:
The system achieves universality by implementing a unified automated XAI prediction process that can handle structured data, semi-structured data, and unstructured text data through the same deep neural network framework. The system uses universal preprocessing and feature extraction mechanisms that adapt to different data types without requiring separate specialized processing pipelines.
Data Source
AI summary
This disclosure relates generally to system and method for determining explainability of machine predicted decisions. Typical explainable AI (XAI) solutions are limited by type of data processed, such as structured, semi-structured and unstructured text. In addition, due to limited automation of the process of explainability, typical systems are cumbersome and time-consuming. The system and method provide an end to end solution for automating the determination of explainability of machine predicted decisions. The XAI process output an absolute relevance score indicative of relevance of the features associated with the prediction which is indicative of percentage relevance/contribution of individual feature. The system further computes relative relevance score of the features by adding up all the features and calculating how much each individual feature is contributing to the total score. The relative relevance scores are utilized for determining explainability of decisions of the prediction.