Process Data Attribution Using Feature-Dependency Sample Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for interpreting model results in data analysis, such as the Shapley value-based approach, are limited in effectively attributing the influence of multiple process features on output results, particularly in complex processes like semiconductor fabrication, where dependencies between features are not adequately addressed.
Innovation Solution
A processor-implemented method generates sample data by modifying reference data based on dependencies between process features, using multiple machine learning models to calculate attributions and confidence, allowing for the identification and adjustment of key process features influencing output results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Shapley value-based attribution analysis is used to interpret model results, then attribution values for multiple features can be calculated, but the dependencies between process features are not adequately addressed
Solution Approach 1:
The patent segments the attribution analysis process into multiple stages: (1) generating sample data by modifying reference data while preserving feature dependencies, (2) training a separate attribution analysis model on this structured sample data, and (3) using the trained model to predict attributions for new data. This segmentation allows the system to handle feature dependencies systematically rather than treating all features independently as in traditional Shapley value approaches.
Solution Approach 2:
The patent performs preliminary actions by pre-processing reference data to generate sample data that explicitly encodes feature dependencies before training the attribution model. This preliminary structuring of data with dependency relationships built-in allows the model to learn and capture complex feature interactions, addressing the limitation of traditional methods that fail to account for such dependencies.
2Adaptability or versatility
If reference data is modified to generate sample data for attribution analysis, then feature dependencies can be captured, but data modification complexity increases
Solution Approach 1:
The patent introduces an intermediary component - a sample data generation module that acts as a mediator between the reference data and the attribution model training process. This intermediary automatically modifies reference data by adjusting feature values while preserving dependency relationships, eliminating the need for manual data modification and reducing overall system complexity despite the sophisticated data transformation required.
Solution Approach 2:
The patent systematically changes data parameters during sample generation by modifying feature values according to predefined dependency rules. Instead of complex structural modifications, the system achieves feature dependency capture through controlled parameter changes in the data, simplifying the modification process while maintaining analytical accuracy.
3Measurement precision
If multiple machine learning models are used to calculate attributions and confidence, then attribution accuracy improves, but computational resources and processing time increase
Solution Approach 1:
The patent performs preliminary action by training the attribution analysis model on pre-generated sample data before actual attribution analysis is needed. This offline training phase allows the model to learn feature dependency patterns and attribution relationships in advance, so that during actual use, attributions can be predicted quickly using the trained model rather than requiring multiple complex calculations for each new data point.
Solution Approach 2:
The patent creates a copied representation of the complex multi-model attribution process by training a single attribution analysis model on sample data generated from multiple reference data variations. This copied model encapsulates the knowledge from multiple sources and can replicate their attribution capabilities more efficiently, reducing the need to run multiple separate models for each analysis task.
Data Source
AI summary
A method and device with process attribution identification are provided. The method may include generating a process result using a first machine learning model provided input data, where the input data incudes feature values corresponding to a plurality of process features, generating sample data by a first modifying of at least a portion of reference data based on dependency between two or more of the plurality of process features, where the reference data includes a plurality of feature values for a reference process result, identifying an attribution of the plurality of process features based on the generated process result and a sample process result generated using the first machine learning model, or a second machine learning model related to the first machine learning model, provided the generated sample data.