Neural Network Model Reshaping for Hardware Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural network models often waste hardware resources and decrease operating speed due to suboptimal hardware utilization, as their characteristics are not optimized for specific hardware.
Innovation Solution
A method to reshape neural network models by determining dedicated hardware information and adjusting the model's parameters, such as weight values and kernel sizes, to optimize execution on specific hardware devices, thereby improving performance and reducing power consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a general neural network model is used without hardware optimization, then the model can be executed on various hardware platforms, but hardware resources are wasted and operating speed decreases
Solution Approach 1:
The patent changes the parameters of the neural network model (such as weight precision, activation function selection, and layer configuration) to match the specific characteristics of the target hardware device. This optimization process transforms a general-purpose model into a hardware-specific model that maximizes operating speed while minimizing hardware resource consumption.
Solution Approach 2:
The patent segments the model optimization process into distinct stages: parsing the input model, determining hardware device information, and generating a reshaped model tailored to the specific hardware. This segmentation allows systematic optimization for different hardware platforms.
2Reliability
If the neural network model is optimized for specific hardware, then operating performance improves, but the model must be reshaped for each hardware device
Solution Approach 1:
The system performs self-service by automatically parsing the input neural network model, analyzing hardware device information, and generating an optimized reshaped model without requiring manual intervention. This automation reduces the complexity of model reshaping while maintaining high operating performance.
Solution Approach 2:
The patent incorporates feedback mechanisms where the system determines hardware device information and uses this information to adjust and generate the reshaped model. This feedback loop ensures the model is properly optimized for the specific hardware while automating the complex reshaping process.
3Productivity
If hardware resources are allocated for neural network execution, then the neural network can be executed, but power consumption increases
Solution Approach 1:
The patent optimizes model parameters such as weight precision and computational complexity to match the hardware capabilities, achieving efficient neural network execution with minimized power consumption. By adjusting these parameters, the system maintains execution capability while reducing energy usage.
Data Source
AI summary
A method of operating a neural network system includes parsing, by a processor, at least one item of information related to a neural network operation from an input neural network model; determining, by the processor, information of at least one dedicated hardware device; and generating, by the processor, a reshaped neural network model by changing information of the input neural network model according to a result of determining the information of the at least one dedicated hardware device such that the reshaped neural network model is tailored for execution by the dedicated hardware device.