AI Explanation Dataset Generation via Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Modern machine learning methods, particularly AI modules, face challenges in explaining their decision-making processes, making it difficult to certify their use in applications like autonomous driving and leading to regulatory issues due to the inability to provide human-understandable explanations for their outputs.
Innovation Solution
A method is developed to generate an explanation dataset by modeling the task of finding explanations as an optimization problem, where a user specifies a target specification and metrics, allowing for the computation of an explanation dataset that provides alternative input data resulting in a desired output, with constraints to limit the solution space and ensure relevance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If AI modules use complex data structures with large numbers of weights and parameters, then the accuracy and capability of the AI module is improved, but the explainability and interpretability of the AI module deteriorates
Solution Approach 1:
The patent segments the explanation task from the complex AI module by introducing a separate explanation generation system. The explanation dataset is created independently through optimization processes, separating the complex prediction function from the simpler explanation generation function. This allows the AI module to maintain its complexity for accuracy while the explanation system remains more interpretable.
Solution Approach 2:
The patent introduces an intermediary explanation generation system that bridges the complex AI module and the user. Instead of directly interpreting the complex weights and parameters, the system creates an intermediate explanation dataset that translates complex decisions into human-understandable alternatives, making the black box accessible without exposing its internal complexity.
2Loss of information
If technical explanations specifying hyperparameters and weights are provided, then the completeness of the explanation is improved, but the usability and understandability for end users deteriorates
Solution Approach 1:
The patent changes the parameters of the explanation from technical (hyperparameters, weights) to practical (alternative input data, counterfactual scenarios). The explanation dataset transforms the representation of information from the internal model parameters to user-relevant features, maintaining completeness while dramatically improving usability.
Solution Approach 2:
The patent creates a copy of the input data structure (explanation dataset) that mirrors the user's original data format and concerns. Instead of presenting internal weights, the system generates alternative versions of the user's own data that would lead to different outcomes, making the explanation immediately relatable and understandable.
3Loss of information
If multiple explanation datasets are generated to provide comprehensive explanations, then the thoroughness of the explanation is improved, but the computational time and resources required deteriorates
Solution Approach 1:
The patent applies partial action by generating a limited number of targeted explanation scenarios rather than exhaustively exploring all possible explanations. The optimization process creates a manageable set of explanation datasets that provide sufficient thoroughness for user understanding without requiring complete enumeration of all possible counterfactuals.
Solution Approach 2:
The patent changes the optimization parameters to balance thoroughness and efficiency. By adjusting the number of explanations to generate and the optimization criteria, the system achieves sufficient explanation quality without excessive computational cost, finding the optimal trade-off point.
Data Source
AI summary
Explaining the decisions of AI modules to a user is difficult.The invention relates to methods for providing an explanation dataset (2) for an AI module (31), the methods comprising:receiving a user dataset (20) which specifies at least one input dataset (21) of an AI module (31), wherein the AI module (31) is adapted to compute an output dataset (3) for the input dataset (21), wherein the user dataset (20) comprises at least one target specification (25) which specifies a value of a data item (26) in an output dataset (3) of the AI module (31);loading at least one optimization task (16) which specifies a specific metric (14) and/or a similarity metric (15);computing at least one solution of the at least one optimization task (16) as an explanation dataset (2) taking the user dataset (20) and the AI module (31) into consideration and applying at least one optimization method (17), wherein the AI module (31) is adapted to compute for the explanation dataset (2) an output dataset (3) which comprises the data item (26) specified by the target specification (25);providing the explanation dataset (2) for the AI module (31).