Neural Network Model Layer Segmentation for Storage Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing systems for managing neural network models face challenges in reducing storage space and network bandwidth usage, as they require storing and transferring multiple versions of pre-trained models, which leads to increased storage demands and high bandwidth consumption.
Innovation Solution
A system with a proxy unit between model servers and client devices that allows only modified layers of neural network models to be uploaded and downloaded, using a processor unit to manage metadata, cache models, and prioritize data transfer based on layer changes and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all versions of neural network models are stored in the model database, then complete model version history is maintained, but storage space consumption increases significantly
Solution Approach 1:
The patent segments the neural network model into multiple independent layers. Instead of storing complete model versions, only the modified layers are stored in the database. Each layer can be independently managed and retrieved, allowing the system to reconstruct complete models by combining stored modified layers with unchanged layers from previous versions.
Solution Approach 2:
The patent extracts only the modified portions (specific layers) of the neural network model from the complete model version and stores only these extracted modifications in the database. This extraction approach eliminates the need to store redundant unchanged portions of the model across different versions.
2Reliability
If complete neural network models are transferred between model server and client devices, then all model updates are delivered, but network bandwidth consumption increases
Solution Approach 1:
The system extracts and transfers only the modified layers of the neural network model between the model server and client devices. The proxy unit identifies which layers have changed and facilitates transfer of only these specific layers, rather than transferring complete model versions, thereby reducing network bandwidth consumption.
Solution Approach 2:
The proxy unit performs preliminary analysis to identify modified layers before initiating the transfer process. By pre-processing the model comparison and identifying only the necessary components that need transfer, the system avoids unnecessary data transmission and optimizes network bandwidth utilization.
3Volume of stationary object
If compression is applied to reduce model size, then storage space is reduced, but compression provides insufficient savings for binary data
Solution Approach 1:
The patent segments the model into layers and stores only the modified segments. This segmentation approach provides more effective space savings than compression because it eliminates entire unchanged layers from storage rather than attempting to compress binary data, achieving superior reduction in storage requirements.
4Reliability
If all model layers are transferred and stored, then complete model versions are maintained, but the transfer time increases substantially
Solution Approach 1:
The system extracts and transfers only the modified layers between server and client, significantly reducing the data volume that needs to be transmitted. This extraction approach maintains model version integrity by ensuring all necessary modifications are captured while minimizing transfer time by excluding unchanged layers from the transfer process.
Solution Approach 2:
The proxy unit performs preliminary comparison to identify modified layers before the transfer operation. This pre-processing step enables the system to prepare only the necessary data for transfer, reducing transfer time while maintaining complete model version integrity through accurate identification of required updates.
Data Source
AI summary
A system is provided. The system includes at least one model server which allows at least one client device to download at least one of pluralities of pre-trained neural network models including pluralities of pre-trained layers in a model database and to upload said models to the model database. Accordingly, the improvement of the system is that a proxy unit, which has at least one processor unit, is provided between said model server and said client devices; said processor unit is configured to realize the steps of accessing the neural network models, in case the client device requests a neural network model, transmitting said neural network model to the client device, in case the client device requests uploading of an edited neural network model, including at least one modified layer where the client device made change, to the model server, uploading only said modified layers to the model database.