Neural Network Structure Search Device
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for searching neural network structures require numerous training iterations, consuming significant computational resources and time, as the recognition accuracy of a neural network structure is unknown until weights are updated through repeated training, making it difficult to find an optimal structure efficiently.
Innovation Solution
A neural network structure search device that trains a model, analyzes the trained structure to identify redundant elements, and generates a new structure by deleting unimportant elements, reducing the number of training times required while maintaining recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If repeated training is performed to evaluate neural network structure accuracy, then recognition accuracy can be determined, but the search time and computational resources increase significantly
Solution Approach 1:
The patent performs importance analysis of network elements before final structure determination. By analyzing which elements contribute most to recognition accuracy during training, the system can identify and retain only critical elements, reducing the number of repeated training iterations needed to evaluate different structures.
Solution Approach 2:
The patent extracts and removes unimportant elements from the neural network structure based on importance analysis. This extraction process identifies redundant elements that do not contribute significantly to recognition accuracy, allowing the system to focus training evaluations only on essential structural components.
2Measurement precision
If repeated training is performed to evaluate neural network structure accuracy, then recognition accuracy can be determined, but computational resources are consumed excessively
Solution Approach 1:
The patent performs importance analysis of network elements before final structure determination. By analyzing which elements contribute most to recognition accuracy during training, the system can identify and retain only critical elements, reducing the number of repeated training iterations needed to evaluate different structures.
Solution Approach 2:
The patent extracts and removes unimportant elements from the neural network structure based on importance analysis. This extraction process identifies redundant elements that do not contribute significantly to recognition accuracy, allowing the system to focus training evaluations only on essential structural components.
3Power
If the number of layers or channels is increased, then recognition processing capacity increases, but recognition accuracy does not necessarily improve
Solution Approach 1:
The patent applies local quality by differentiating the importance of individual elements (layers, channels, or specific network components) rather than treating all elements uniformly. Elements are pruned or retained based on their specific contribution to recognition accuracy, allowing the network to maintain appropriate complexity only where needed.
Solution Approach 2:
The patent extracts and removes unimportant elements from the neural network structure based on importance analysis. This extraction process identifies redundant elements that do not contribute significantly to recognition accuracy, allowing the system to focus training evaluations only on essential structural components.
Data Source
AI summary
A neural network structure search device includes a training unit which trains a neural network model with a first neural network structure using a data set for training, a first generation unit which generates analysis information indicating an importance of each of elements comprising the first neural network structure by analyzing a trained model generated from the neural network model by the training using the data set for training, an identifying unit which identifies an element in the first neural network structure whose importance is lower than a predetermined value using the generated analysis information, and a second generation unit which generates a second neural network structure based on the first neural network structure by deleting the identified element from the first neural network structure.