Neural Network Structure Search Using CPU Utilization Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural network models incur high CPU utilization and resource consumption during operation, limiting their practical application due to neglecting CPU utilization in the model searching process.
Innovation Solution
A method and apparatus that determine a neural network model structure by predicting runtime CPU utilization using a CPU utilization prediction model, ensuring the model meets a preset requirement, thereby reducing resource consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If automatic neural architecture search is conducted considering network performance and parameters like size and computing amount, then model performance is improved, but runtime CPU utilization increases
Solution Approach 1:
The patent applies preliminary action by predicting runtime CPU utilization during the model search phase using a pre-built CPU utilization prediction model. This allows the system to identify and select models with low predicted CPU utilization before actual deployment, thereby avoiding high runtime resource consumption while maintaining performance requirements.
Solution Approach 2:
The patent changes the optimization parameters from only considering model performance metrics to including predicted CPU utilization as a key parameter. By incorporating CPU utilization prediction into the model selection criteria, the system identifies models that balance both performance and resource efficiency, thus resolving the contradiction between performance and runtime resource consumption.
2Reliability
If large scale network architecture is used to improve model performance, then accuracy is improved, but resource consumption increases
Solution Approach 1:
The system performs preliminary evaluation of resource consumption by predicting CPU utilization for each candidate model during the architecture search process. This allows filtering out models that would consume excessive resources before deployment, enabling selection of models with optimal balance between accuracy and resource efficiency.
Solution Approach 2:
The patent modifies the model selection parameters to include predicted CPU utilization alongside accuracy metrics. This multi-parameter optimization approach enables identification of models that achieve acceptable accuracy with minimized resource consumption, resolving the trade-off between model scale and resource usage.
Data Source
AI summary
The embodiments of the present disclosure provide a neural network model structure determining method and apparatus, a device, a medium and a product. The neural network model structure determining method includes: determining, based on a preset neural network model architecture search algorithm, at least one candidate neural network model; predicting, based on a preset Central Processing Unit (CPU) utilization prediction model, a runtime CPU utilization for each candidate neural network model, so as to obtain a predicted value of CPU utilization; and determining, based on at least one predicted value of CPU utilization, a target neural network model structure among structures of the at least one candidate neural network model.