AI Processor Parameter Classification for Hardware Size Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep learning-based image processing technologies, such as super resolution, require numerous multipliers for convolution operations, leading to hardware size, cost, and power consumption burdens due to the complexity of neural network operations.
Innovation Solution
An electronic apparatus and control method that identify whether parameters in an artificial intelligence model's layers are fixed or variable, optimizing operations by using fewer multipliers and simplifying hardware, particularly by implementing operation modules with a reduced number of multipliers or only adders, based on parameter identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If deep learning-based image processing is implemented with numerous multipliers for convolution operations, then image processing capability is improved, but hardware size and power consumption increase
Solution Approach 1:
The patent changes the parameter type identification mechanism, classifying parameters as fixed or variable to determine the appropriate operation module. This parameter-based classification allows the system to optimize hardware resource allocation by matching operation types with suitable computational units, thereby reducing overall hardware size while maintaining processing capability.
Solution Approach 2:
The patent segments the convolution operation into two distinct types: fixed parameter operations and variable parameter operations. By dividing the operation modules into separate first operation modules (for fixed parameters) and second operation modules (for variable parameters), the system can optimize each segment independently, reducing the total hardware footprint required for image processing.
2Productivity
If deep learning-based image processing is implemented with numerous multipliers for convolution operations, then image processing capability is improved, but power consumption increases
Solution Approach 1:
The patent introduces parameter type identification that changes the operational mode based on whether parameters are fixed or variable. This dynamic parameter-based control allows the system to select energy-efficient operation paths, using first operation modules for fixed parameters which require fewer computational resources and thus consume less power compared to universal multiplier-based approaches.
Solution Approach 2:
By segmenting operation modules into specialized first operation modules and second operation modules, the patent enables energy-optimized processing. Fixed parameter operations routed to first operation modules consume less power than variable parameter operations, thereby reducing overall power consumption while maintaining image processing capability.
3Ease of manufacture
If fixed parameters are identified and provided to first operation modules instead of second operation modules, then hardware resource utilization is improved, but operation complexity increases
Solution Approach 1:
The patent uses parameter type identification as a control mechanism that simplifies hardware resource allocation. By changing the operational parameter from a generic computation mode to a specific fixed/variable parameter classification, the system automatically routes operations to appropriate modules, improving resource utilization while the classification logic itself remains relatively simple.
Solution Approach 2:
The patent performs preliminary classification of parameters as fixed or variable before executing convolution operations. This preliminary action of parameter identification enables subsequent operation modules to be pre-configured for their specific function, improving hardware resource utilization by matching operations with specialized modules rather than using a universal approach.
Data Source
AI summary
An electronic apparatus is provided. The electronic apparatus includes: a memory storing information on an artificial intelligence model including a plurality of layers, and a processor configured to acquire an output image based on processing an input image using the artificial intelligence model. The processor is configured to: identify whether a parameter used in any one layer among the plurality of layers is a fixed parameter or a variable parameter, and provide the parameter to a first operation module or a second operation module included in of the artificial intelligence model of the artificial intelligence model any one layer based on the identification and perform an operation between output data of a previous layer and the parameter.