ML Model Retraining via Surrogate Feature Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI/ML models struggle to incorporate new features that appear during the testing or deployment phase, leading to performance degradation as these models are not trained to identify new features, resulting in a waste of valuable data resources.
Innovation Solution
A method and system for re-training a Machine Learning model by predicting new features using a surrogate model, which generates predicted values for new features based on the testing feature set and incorporates these values into the training feature set, allowing the model to adapt to new features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a new feature is added during the testing or deployment phase, then the model's predictive capability is improved, but the model cannot utilize the new feature because it was not trained on it
Solution Approach 1:
The patent applies preliminary action by training a surrogate model to predict the values of new features before the main model can utilize them. The surrogate model is trained on the new feature values that appear during testing, and then uses these predictions to create an augmented training dataset that includes the new features. This allows the main model to be retrained with the new features in advance, resolving the contradiction between adaptability and reliability.
2Adaptability or versatility
If the model is re-trained from scratch with new data including new features, then the model can learn new features, but valuable historical data resources are wasted
Solution Approach 1:
The patent applies merging by combining the historical training dataset with newly generated synthetic data that includes the new features. The surrogate model predicts new feature values for historical data samples, and these predicted values are merged with the original training data to create an augmented dataset. This allows the model to learn new features while preserving and utilizing all historical data resources, eliminating the waste associated with retraining from scratch.
Solution Approach 2:
The patent applies copying by creating synthetic copies of historical data samples with predicted new feature values. Instead of discarding historical data, the system generates augmented versions of these samples that include the new features through the surrogate model's predictions. These copied and augmented samples are then used to retrain the model, allowing learning of new features while fully utilizing the value of historical data resources.
3Reliability
If extensive re-training is performed to incorporate new features, then the model performance is improved, but the time and computational resources required increase significantly
Solution Approach 1:
The patent applies segmentation by dividing the re-training process into two separate stages: first training a surrogate model to predict new feature values, and then using those predictions to augment and retrain the main model. This segmentation allows the system to efficiently handle the complexity of incorporating new features without requiring extensive simultaneous processing, thereby reducing the overall time and computational resources needed compared to traditional extensive re-training approaches.
Data Source
AI summary
Methods and systems for re-training a Machine Learning (ML) model using predicted features from a training dataset are disclosed. A method performed by a server system includes accessing a training feature set and a testing feature set from a database. In response to identifying an inclusion of at least one new feature in the testing feature set, the method includes training a surrogate ML model to predict a value for the new feature based on the testing feature set and determining, by the surrogate ML model, a predicted value for the new feature for each training data sample in a training dataset based on the training feature set. The method further includes generating a new training feature set for each training data sample based on the predicted value and the training feature set. The method includes re-training the ML model based on the new training feature for each data sample.