Model Explanation via Differential Credit Assignment Sampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning systems face challenges in providing insight into their operations, making it difficult to assess safety, compliance with regulations, and understanding feature importance, especially when generating explanation information for large datasets, which can be computationally expensive and time-consuming.
Innovation Solution
A method that uses sampling techniques to generate explanation information by performing a credit assignment process, allowing for the evaluation of machine learning systems by sampling inputs until convergence criteria are met, thereby reducing computational burden and improving performance without compromising accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If common methods for model explanation are used, then explanation information can be generated, but computational cost and time consumption increase significantly when processing large datasets
Solution Approach 1:
The patent segments the large dataset into smaller batches or samples for processing. Instead of computing explanation information for the entire dataset at once, the system processes data in manageable chunks, reducing the computational burden and time required while maintaining explanation accuracy through systematic sampling strategies.
Solution Approach 2:
The patent applies partial action by processing a representative subset or sample of the data rather than the complete dataset. By strategically selecting and processing only the necessary portion of data that captures the essential patterns and relationships, the system achieves adequate explanation accuracy with significantly reduced computational resources and time.
2Reliability
If complete dataset processing is performed for model explanation, then comprehensive explanation information is obtained, but computational expense increases
Solution Approach 1:
The patent performs preliminary actions by pre-processing and preparing data samples before the main explanation computation. This includes data normalization, feature selection, and sample stratification that are done in advance, allowing the main computational process to work more efficiently with pre-prepared data, thereby reducing overall energy consumption while maintaining reliability.
Solution Approach 2:
The patent changes key parameters such as sample size, sampling rate, and processing batch size to optimize the balance between explanation reliability and computational energy consumption. By dynamically adjusting these parameters based on data characteristics and resource availability, the system achieves reliable explanations with minimized energy usage.
3Loss of information
If detailed explanation information is generated for all features, then comprehensive insight is provided, but processing complexity increases
Solution Approach 1:
The patent extracts only the most relevant and important feature explanation information rather than computing explanations for all features equally. By identifying and extracting key features that contribute most to model predictions, the system provides comprehensive insight into critical decision factors while avoiding the complexity of processing and presenting information for every single feature.
Solution Approach 2:
The patent applies local quality by providing different levels of explanation detail for different features based on their importance and relevance. Critical features receive detailed explanation information, while less important features receive simplified or aggregated explanations, optimizing the balance between information completeness and processing complexity.
Data Source
AI summary
Systems and methods for model explanation are disclosed. In one embodiment, the disclosed process determines a score based on a scoring function and a plurality of values associated with a plurality of features of a denied credit applicant. (e.g., credit score of 550, no loans repaid, etc.). The process then determines a score of an approved credit applicant. (e.g., credit score of 750, 3 loans repaid, etc.). A next differential credit assignment associated with the current denied/approved pair is then calculated. If a convergence stopping criteria, (e.g., current accuracy>99% based on a statistical t-distribution) is not satisfied, the process repeats for a different approved credit applicant. When the convergence stopping criteria is satisfied, explanation information is generated. For example, the explanation information may include an adverse action reason code, fairness metric, disparate impact metric, human readable text, feature importance metric, credit value, and/or an importance rank.