Neural Network Device Adaptive Class Combination Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current neural networks face limitations in improving performance as increasing the number of identifiable classes often decreases identification accuracy, making it difficult to enhance both versatility and accuracy simultaneously.
Innovation Solution
A neural network device and method that adaptively selects and configures class combinations based on usage situations by varying the weights and inputs of specific layers, allowing for accurate identification of required classes with reduced storage capacity, similar to having multiple networks with a small storage footprint.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the number of classes that a neural network can identify is increased, then the versatility of the neural network is improved, but the identification accuracy deteriorates
Solution Approach 1:
The patent divides the neural network into multiple independent modules, each specialized for identifying a specific subset of classes. Instead of one large network handling all N classes, the system segments the classification task across multiple smaller networks, where each network is optimized for its specific class subset, thereby maintaining high accuracy while collectively covering all classes.
Solution Approach 2:
The patent introduces dynamic selection mechanisms that allow the system to adaptively choose which neural network module to use based on the input data characteristics. This dynamic approach enables the system to switch between different specialized networks, optimizing accuracy for each specific classification task while maintaining overall versatility across all classes.
2Measurement precision
If multiple neural networks with different class combinations are created, then the identification accuracy for specific classes is improved, but the storage capacity requirement increases
Solution Approach 1:
The patent merges multiple specialized neural network modules into a single integrated system that shares common computational resources, such as feature extraction layers. By combining these modules with shared components, the system reduces redundant storage requirements while maintaining the accuracy benefits of specialized networks for different class combinations.
Solution Approach 2:
The patent designs neural network modules with universal components that can serve multiple functions across different class combinations. Shared layers and parameters are designed to be reusable across multiple specialized networks, allowing the system to maintain multiple specialized models without proportionally increasing storage capacity, as each module contributes to multiple classification tasks.
Data Source
AI summary
The disclosed embodiment provides a neural network device and an output handling method thereof, that selects at least one class combination of different numbers and types from N (where N is a natural number) classes designated for input data, sets, according to each class combination selected in a neural network module including a plurality of layers, at least one layer among the layers of the neural network module as a changeable variable layer, and outputs a result of performing a neural network operation on the input data by changing the variable layer set according to the selected class combination, and a weight of the variable layer. The neural network device and output handling method thereof adaptively determines the number and type of classes to be identified by varying them in various combinations depending on the usage situation, and identifies classes according to the determined class combination.