Review-Based Machine Learning Model Retraining
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models require extensive time and computational resources to retrain due to the need for large amounts of new raw data, leading to delays in model updates when errors exceed a threshold.
Innovation Solution
A review-based machine learning approach that reuses existing features and labels to generate new models, allowing for conditional model updates using a trigger mechanism based on timer expiration, label thresholds, or performance errors, thereby reducing the need for new feature generation and data preparation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a machine learning model is retrained using large amounts of new raw data, then model accuracy can be maintained or improved, but the time required for model updates increases significantly
Solution Approach 1:
The patent applies preliminary action by pre-computing and storing feature representations of training data in advance. When model retraining is needed, these pre-computed features are reused instead of processing raw data again, significantly reducing the time required for model updates while maintaining accuracy through the use of high-quality feature representations.
Solution Approach 2:
The patent discards the time-consuming raw data processing step by recovering and reusing previously computed feature representations. Instead of discarding the computational effort already invested in feature extraction, the system recovers these features for reuse in model retraining, eliminating redundant computation while preserving data quality.
2Reliability
If new raw data is collected and prepared for model retraining, then model performance can be improved, but computational resources and processing time are consumed
Solution Approach 1:
The system recovers previously computed feature representations instead of discarding them after initial model training. These recovered features are reused for model retraining, eliminating the need to reprocess raw data and significantly reducing computational resource consumption while maintaining model performance.
Solution Approach 2:
Computational resources are saved by performing feature extraction in advance as a preliminary action. The pre-computed features are stored and reused when model retraining is needed, avoiding the energy-intensive process of重新 processing raw data through complex feature extraction algorithms.
3Manufacturing precision
If extensive data preparation is performed before model retraining, then training quality is improved, but the delay before retraining can begin increases
Solution Approach 1:
Data preparation is performed in advance as a preliminary action, with feature representations being computed and stored before model training is needed. This eliminates the delay before retraining can begin, as the high-quality feature data is already prepared and available for immediate use in model retraining.
Solution Approach 2:
The system recovers pre-prepared feature data instead of performing extensive data preparation at the time of retraining. This eliminates the time delay associated with data preparation while maintaining training quality, as the same high-quality features are reused without needing to be regenerated.
Data Source
AI summary
An apparatus and method are provided for review-based machine learning. Included are a non-transitory memory storing instructions and one or more processors in communication with the non-transitory memory. The one or more processors execute the instructions to receive first data, generate a plurality of first features based on the first data, and identify a first set of labels for the first data. A first model is trained using the first features and the first set of labels. The first model is reviewed to generate a second model, by receiving a second set of labels for the first data, and reusing the first features with the second set of labels in connection with training the second model.