Processor Neural Network Reconstruction for Shared Layer Storage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing use of artificial neural networks (ANNs) in electrical devices leads to storage space limitations and performance degradation due to the need to store multiple networks, which often have overlapping layer groups.
Innovation Solution
A processor is used to analyze the structure of multiple neural networks, identify sharing layer groups, and reconstruct them to include a common layer group, storing this group only once in the device's memory, thereby reducing storage requirements and maintaining performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple neural networks are stored to handle different tasks, then task versatility is improved, but storage space requirement increases
Solution Approach 1:
The patent merges overlapping layer groups from multiple neural networks into a single shared layer group. By analyzing the structure of multiple neural networks and identifying common layers, the system reconstructs the networks to share these layers, thereby reducing total storage space while maintaining the ability to perform multiple tasks.
Solution Approach 2:
The shared layer group is designed to be universal and can be used by multiple different neural networks for different tasks. This multi-functional design allows a single layer group to serve multiple purposes across different network architectures, reducing redundancy and storage requirements.
2Adaptability or versatility
If multiple neural networks with overlapping layers are stored, then functional capability is improved, but device performance degrades due to storage limitations
Solution Approach 1:
By merging overlapping layers into a shared structure, the system reduces storage footprint which was causing performance degradation. The shared layer group maintains functional capability while the reduced storage requirements improve device performance by preventing storage-related bottlenecks.
3Quantity of substance
If neural network layers are reconstructed to share common groups, then storage space is reduced, but calculation speed may be affected
Solution Approach 1:
The patent segments neural network layers into sharing layer groups and non-sharing layers. This segmentation allows efficient storage of shared layers while maintaining the ability to quickly access and calculate with them. The segmented structure enables optimized memory access patterns that preserve calculation speed despite the shared architecture.
Data Source
AI summary
An operation method of a processor configured to operate based on an artificial neural network including a first neural network and a second neural network is provided. The operation method includes: analyzing similarity of a structure of the first neural network and a structure of the second neural network, the first neural network including a plurality of first neural network layers and the second neural network including a plurality of second neural network layers; selecting, from the plurality of first neural network layers and the plurality of second neural network layers, sharing layers capable of being commonly used, based on a result of the analysis; and reconstructing the structure of the first neural network or the structure of the second neural network based on the sharing layers.