ML Model Reuse via Data Similarity Assessment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The high computational and storage costs associated with continuously training and retraining machine learning models, especially in edge computing devices and sensors, where data patterns often repeat and models become redundant.
Innovation Solution
Implementing a system that reuses trained machine learning models when the similarity between the dataset used to train the initial model and a later-received dataset exceeds a certain threshold, thereby reducing the need for constant retraining and optimizing data storage and processing resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning models are continuously trained and retrained on new data, then model accuracy is maintained or improved, but computational resources and storage costs increase significantly
Solution Approach 1:
The patent creates simplified copies or representations of training data characteristics (similarity metrics) that can be compared without full retraining. Instead of copying and retraining entire models, the system copies key data characteristics and uses them to determine model reusability, significantly reducing computational overhead while maintaining accuracy.
Solution Approach 2:
The system performs preliminary similarity assessment between new data and existing training data before committing to full model retraining. This preliminary action filters out cases where retraining is unnecessary, saving substantial computational resources while ensuring accuracy is maintained when needed.
2Reliability
If machine learning models are continuously retrained, then model accuracy is maintained, but storage space for multiple model versions increases
Solution Approach 1:
The system stores compact representations of data similarity metrics and model metadata rather than full copies of multiple model versions. This copying approach preserves the ability to assess model applicability while dramatically reducing storage requirements.
Solution Approach 2:
The system discards redundant full model copies when similarity thresholds indicate existing models remain applicable. Computational resources and storage are recovered by avoiding unnecessary model retention, while the ability to recover and reuse models is maintained through similarity tracking.
3Adaptability or versatility
If data is stored for continuous model training, then model retraining flexibility is maintained, but data storage costs increase by up to 70%
Solution Approach 1:
The system stores compact copies of data similarity metrics and characteristics rather than complete datasets. These copied metrics retain the essential information needed to assess model applicability and maintain retraining flexibility while reducing storage requirements by up to 70%.
Solution Approach 2:
The system extracts and stores only the critical similarity metrics and data characteristics needed for model selection, leaving out redundant full dataset storage. This extraction maintains the ability to make informed retraining decisions while significantly reducing storage costs.
4Adaptability or versatility
If model retraining is performed frequently, then models adapt to new data patterns, but processing time and computational overhead increase
Solution Approach 1:
The system performs preliminary similarity checks before initiating full model retraining. This preliminary assessment quickly identifies cases where existing models remain suitable, avoiding time-consuming retraining operations while ensuring adaptation occurs when genuinely needed.
Solution Approach 2:
The system skips unnecessary full retraining cycles by using similarity thresholds to identify cases where existing models remain applicable. This skipping mechanism rushes through redundant processing steps while maintaining model adaptability through targeted retraining when data patterns genuinely change.
Data Source
AI summary
Systems and methods are provided for reusing machine learning models. For example, the applicability of prior models may be compared using one or more assessment values, including a similarity threshold and/or an accuracy threshold. The similarity threshold may identify a similarity of data between a first data set used to generate a first model and a new data set that is received by the system. When the similarity between these two data sets is exceeded, the system may reuse a model with the highest similarity value. When an accuracy value of the data set does not exceed an accuracy threshold, the system may initiate a retraining process to generate a second ML model associated with the second data.