Ultimate Reason Code Generation Through Submodel Retraining
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating reason codes for complex non-linear models in big data applications, such as neural networks, rely on single models that assume score consistency without retraining, which is not always valid, leading to inconsistent results.
Innovation Solution
A system and method that trains a base model with multiple reason codes, trains subsequent models with subsets of these codes removed, and identifies the reason code with the largest score drop as the ultimate reason code, ensuring consistency with probability of the target.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a single base model is used to compute reason codes by modifying input variables, then the computation is simple and fast, but the score consistency assumption does not hold and results are unreliable
Solution Approach 1:
The patent segments the single model approach into multiple sub-models, each trained with a specific reason code subset. This segmentation allows each sub-model to maintain score consistency for its designated reason codes while collectively covering all reason codes, thus resolving the reliability issue without requiring a completely complex new system architecture.
Solution Approach 2:
The patent applies preliminary action by pre-training multiple sub-models with different reason code subsets before actual scoring. This preliminary training ensures that when reason codes are needed, the pre-trained sub-models can provide consistent scores directly without requiring retraining, thus maintaining reliability while avoiding complexity during runtime.
2Reliability
If multiple sub-models are trained with different reason code subsets, then score consistency is maintained, but the training process becomes more complex
Solution Approach 1:
The training process is segmented into multiple independent sub-model training tasks, each handling a specific reason code subset. This segmentation makes the overall complex training process more manageable by breaking it into smaller, independent units that can be trained separately with standard techniques.
Solution Approach 2:
Each sub-model is trained with a partial subset of reason codes rather than all reason codes. This partial action approach reduces the training complexity for each individual sub-model while collectively achieving complete reason code coverage through the ensemble of sub-models.
3Productivity
If reason codes are generated by derivative computation on a single model, then the process is fast, but the results may be inconsistent with the target probability
Solution Approach 1:
The patent creates multiple copies (sub-models) of the base model, each specialized for specific reason codes. Instead of computing derivatives on a single model, the system uses these copied sub-models to directly predict scores, providing both speed and accuracy by leveraging the pre-trained capabilities of each sub-model for its designated reason codes.
Data Source
AI summary
A system and method for generating ultimate reason codes for computer models is provided. The system for generating ultimate reason codes for computer models comprising a computer system for receiving a data set, and an ultimate reason code generation engine stored on the computer system which, when executed by the computer system, causes the computer system to train a base model with a plurality of reason codes, wherein each reason code includes one or more variables, each of which belongs to only one reason code, train a subsequent model using a subset of the plurality of reason codes, determine whether a high score exists in the base model, determine a scored difference if a high score exists in the base model, and designate a reason code having a largest drop of score as an ultimate reason code.