Machine Learning Model Compression via Sub-Model Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models have grown in size and complexity, leading to increased storage needs and processing resources, necessitating a method to compress these models without degrading their performance.
Innovation Solution
A computing system that performs lossless and lossy compression of machine learning models by removing sub-models based on performance and size metrics, adding them back if performance thresholds are not met, to maintain model performance while reducing size.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are expanded to account for larger numbers of distinctions and conditions, then prediction accuracy is improved, but model size and processing resource requirements increase
Solution Approach 1:
The patent segments the machine learning model into multiple sub-models that can be independently evaluated and selectively removed. Each sub-model processes a portion of the data or handles a specific aspect of the prediction task, allowing the system to divide the large complex model into manageable units for compression analysis
Solution Approach 2:
The patent extracts and removes redundant or less important sub-models from the ensemble based on performance metrics and clustering analysis. By identifying and taking out unnecessary components, the system reduces overall model size while maintaining the predictive capabilities provided by the remaining essential sub-models
2Measurement precision
If machine learning models are expanded to account for larger numbers of distinctions and conditions, then prediction accuracy is improved, but processing resource consumption increases
Solution Approach 1:
The patent extracts and removes redundant or less important sub-models from the ensemble based on performance metrics and clustering analysis. By identifying and taking out unnecessary components, the system reduces overall model size while maintaining the predictive capabilities provided by the remaining essential sub-models
Solution Approach 2:
The patent changes the parameter of model complexity by selectively removing sub-models based on performance thresholds and clustering results. This parameter change allows the system to operate at an optimal point between accuracy and resource consumption, adjusting the effective model size to match available processing resources
3Device complexity
If sub-models are removed from the machine learning model to reduce size, then model size is reduced, but performance may degrade below acceptable thresholds
Solution Approach 1:
The patent implements feedback by evaluating the performance of the compressed model after sub-model removal and comparing it against a performance threshold. If performance degradation is detected, the system provides feedback to adjust the compression strategy, either by retaining additional sub-models or refining the clustering and selection process to maintain acceptable performance levels
Solution Approach 2:
The patent uses parameter changes by adjusting the performance threshold and clustering parameters to find the optimal balance between model size reduction and performance maintenance. By varying these parameters, the system can explore different compression levels and select the configuration that achieves maximum size reduction while staying above the acceptable performance threshold
Data Source
AI summary
A method includes determining a plurality of performance metrics for a plurality of sub-models forming a first machine learning model and clustering the plurality of sub-models based on the plurality of performance metrics to produce a plurality of clusters of sub-models. The method also includes removing, from the first machine learning model, sub-models assigned to a first cluster of the plurality of clusters to produce a second machine learning model formed by the sub-models remaining in the first machine learning model and in response to determining that a performance of the second machine learning model is below a performance threshold, adding a subset of the removed sub-models to the second machine learning model to produce a third machine learning model. The method further includes, in response to determining that a performance of the third machine learning model meets the performance threshold, selecting the third machine learning model to be applied.