Neural Network Variable Deprecation via Preliminary Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models with large numbers of variables face accuracy issues and maintenance challenges due to variable deprecation, leading to costly retraining and inconsistent predictions.
Innovation Solution
Implementing neural networks and decision tree models that self-compensate for deprecated variables by dropping likely-to-be-deprecated variables during training and adjusting predictions using dropout and pruning techniques, respectively, without the need for retraining or forming new models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If variables are deprecated from models to maintain up-to-date information, then model relevance is improved, but prediction accuracy deteriorates due to loss of historical patterns
Solution Approach 1:
The system performs preliminary actions by training multiple versions of the model with different variable sets before actual prediction. Historical models containing deprecated variables are retained and queried when needed, allowing the system to compensate for variable deprecation without losing prediction accuracy.
Solution Approach 2:
A model selection module acts as an intermediary between the variable deprecation process and prediction generation. It selects appropriate historical models based on the current variable set, enabling accurate predictions even when variables are deprecated by matching the right historical model to the current data.
2Measurement precision
If models are retrained when variables are deprecated to maintain accuracy, then prediction accuracy is improved, but computational cost and time increase
Solution Approach 1:
The system performs preliminary training of multiple model versions with different variable sets in advance. When variable deprecation occurs, these pre-trained historical models are immediately available for query without requiring time-consuming retraining operations.
Solution Approach 2:
Instead of retraining the entire model from scratch, the system creates copies of historical models that were trained with different variable sets. These copies are stored and selectively deployed based on current needs, avoiding the time cost of complete model retraining.
3Measurement precision
If all variables are maintained in models to preserve historical data, then prediction accuracy is improved, but model maintenance complexity increases
Solution Approach 1:
The system segments the model into multiple historical versions, each trained with specific variable sets. This segmentation allows individual models to be maintained with relevant variables while others are archived, reducing overall maintenance complexity compared to maintaining all variables across all models.
Solution Approach 2:
The model selection module serves as an intermediary that manages complexity by automatically selecting the appropriate historical model based on current variables. This abstraction layer hides the complexity of multiple model versions from the prediction process, simplifying maintenance operations.
4Adaptability or versatility
If frequent model updates are performed to reflect variable changes, then model relevance is improved, but system stability deteriorates due to inconsistent predictions
Solution Approach 1:
The system performs preliminary training of multiple stable model versions with different variable configurations. These pre-trained models provide stable predictions by relying on historical data patterns rather than continuous updates, maintaining consistency while adapting to variable changes through selective model deployment.
Data Source
AI summary
Various techniques for determining risk assessment predictions and decisions are disclosed. Certain disclosed techniques include the implementation of neural network models in determining predictions of risk for an operation based on an input dataset. The disclosed techniques include training the neural network models to compensate for deprecation of variables from the input dataset. The neural network models may be trained to be robust in view of deprecated variables by dropping variables from the input space during training of the neural network models.