Dual-Subsystem AI Architecture for Network-Isolated Model Execution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computing systems face challenges in maintaining data confidentiality, availability, and integrity, particularly in scenarios where hardware and machine learning models are accessible by networks, leading to unintended data sharing and exposure.
Innovation Solution
A computing system with a dual-subsystem architecture, comprising a public subsystem for network connectivity and a private subsystem for isolated AI operations, along with a power adapter that selectively powers one subsystem at a time, ensuring data confidentiality and integrity by isolating the private subsystem from network connections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the computing system connects to a network for data access and model execution, then productivity and ease of operation are improved, but data confidentiality and integrity deteriorate due to unintended data sharing and exposure
Solution Approach 1:
The computing system is divided into two separate subsystems: a public subsystem with network connectivity for data input and output, and a private subsystem isolated from the network for executing AI models. This segmentation ensures that sensitive AI operations occur in an isolated environment while still allowing the system to access network resources when needed.
Solution Approach 2:
A memory storage device acts as an intermediary between the public and private subsystems. It temporarily stores AI model data and parameters, allowing the public subsystem to provide input data and receive output results without direct network access to the private subsystem where models are executed. This intermediary mechanism enables controlled data transfer while maintaining isolation.
2Reliability
If the private subsystem is isolated from the network to ensure data security, then data confidentiality is improved, but ease of operation and data accessibility worsen
Solution Approach 1:
The memory storage device serves as a mediator that bridges the isolated private subsystem and the network-connected public subsystem. It allows operators to load data into the private subsystem through the public subsystem without compromising the isolation barrier, thus maintaining both security and operational convenience.
Solution Approach 2:
Data and AI model parameters are pre-loaded into the private subsystem's memory storage device through the public subsystem before isolation is enforced. This preliminary action ensures that the private subsystem has all necessary data available for execution without needing network access during operation, maintaining both security and operational efficiency.
3Ease of operation
If network access is provided for data input and output, then ease of operation is improved, but the risk of unintended data sharing and exposure increases
Solution Approach 1:
The system segments network access functionality from AI model execution functionality. The public subsystem handles all network communications for data input and output, while the private subsystem executes models in isolation. This segmentation prevents unintended data sharing by ensuring that sensitive model operations never directly access the network.
Solution Approach 2:
The memory storage device and public subsystem act as intermediaries that control all data flow to and from the network. Data is transferred through controlled interfaces rather than direct network access by the private subsystem, preventing unauthorized or unintended data exposure while maintaining ease of operation for legitimate data input and output.
Data Source
AI summary
A device including a public subsystem comprising a first processor set of processors and a first memory set of memories storing first instructions for a first operating system. The device including a private subsystem comprising a second processor set of processors and a second memory set of memories storing second instructions for a second operating system. The device including a first interface communicatively coupled with the public subsystem and a second interface communicatively coupled with the private subsystem, each interface configured to interconnect the subsystems. The public subsystem configured to present information on a display and connect with a network. The private subsystem configured to present information on the display, incapable of connecting with the network, and configured to receive, from the public subsystem via the second interface, artificial intelligence data and execute, locally by the second processor set, an artificial intelligence model based on the artificial intelligence data.