Local Intelligence Storage for AI Model Management
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI services that utilize cloud and edge computing architectures face inefficiencies in managing deep learning models, as these models become dependent on application containers, leading to constraints in utilization and management.
Innovation Solution
A cloud and edge-based computing device with a local intelligence storage and a local intelligence/model management unit that allows deep learning models to be stored and managed independently from containers, enabling efficient installation, update, and deployment of models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If deep learning models are stored within application containers, then models can be deployed with applications, but model management becomes constrained and inefficient
Solution Approach 1:
The patent separates deep learning model storage from application container storage by introducing a dedicated local intelligence storage unit. This segmentation allows models to be managed independently from applications, enabling flexible model updates, version control, and sharing without requiring application redeployment, thus resolving the management constraints while maintaining deployment capability
Solution Approach 2:
The patent introduces a model management unit as an intermediary component that mediates between the local intelligence storage and application containers. This intermediary handles model installation, updating, and version management, simplifying the complexity of direct model-container coupling while maintaining the ability to deploy models with applications
2Productivity
If deep learning models are stored within application containers, then models are deployed with applications, but updates and version changes cause service delays
Solution Approach 1:
The patent enables preliminary model preparation and validation in the local intelligence storage before deployment to containers. Models can be pre-processed, validated, and staged for deployment without interrupting running applications, reducing service delays during model updates and version changes while improving overall deployment efficiency
3Adaptability or versatility
If deep learning models are stored within application containers, then models are bundled with applications, but model sharing and utilization are constrained
Solution Approach 1:
The patent creates a universal local intelligence storage that serves multiple functions: storing models for multiple applications, enabling model sharing across different containers, supporting version control, and facilitating model updates. This multi-functional storage improves model utilization and sharing while maintaining ease of deployment through standardized access interfaces
Data Source
AI summary
Proposed is a computing device for operating an artificial intelligence (AI) service. The computing device may include a local intelligence storage, in which a deep learning model for AI applications based on a container is stored. The computing device may also include a local intelligence/model management unit configured to scan the local intelligence storage to identify information on the deep learning model installed in the computing device and provide the identified information on the deep learning model through a network interface. The local intelligence/model management unit may manage the deep learning model to be stored on the local intelligence storage that is independent from the container.