Secure AI I/O Shape Changing for Always-On Mobile Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Always-on artificial intelligence (AI) applications in mobile devices face challenges in managing power consumption and securing sensitive data, particularly in ambient intelligence systems where continuous sensing tasks are resource-intensive and vulnerable to attacks.
Innovation Solution
Implementing a secure processor architecture with a trusted execution environment (TEE) and firewalls to protect AI models, coupled with an IO verifier and pre-fire module to manage input/output changes and ensure secure execution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If always-on AI security applications are implemented in mobile devices, then security protection is improved, but power consumption increases
Solution Approach 1:
The system segments the AI security application into a dedicated always-on AI application instance that operates independently from the main application. This segmented architecture allows the security function to run continuously with minimal resource consumption, while the main application can be suspended or reduced to background mode, thus resolving the contradiction between continuous security protection and power consumption.
Solution Approach 2:
The always-on AI application is designed to serve itself by autonomously managing its own execution state, power consumption, and security functions without requiring continuous input from the main application or user. This self-service capability enables the system to maintain security protection while minimizing power consumption through intelligent self-regulation.
2Adaptability or versatility
If flexible IO shape changing is implemented in AI models, then adaptability is improved, but security vulnerability increases
Solution Approach 1:
The system introduces an intermediary verification mechanism between the IO shape changing operation and the AI model execution. This intermediary component validates and authorizes IO shape changes before they are applied to the AI model, thereby enabling flexible adaptability while preventing unauthorized or malicious modifications that would create security vulnerabilities.
Solution Approach 2:
The system performs preliminary verification and authorization actions before allowing IO shape changing to occur. By checking and validating IO shape changes in advance against predefined security policies and model requirements, the system enables flexible adaptation while preventing harmful factors from compromising security.
3Reliability
If secure processor architecture with TEE is implemented, then data protection is improved, but device complexity increases
Solution Approach 1:
The system implements a nested architecture where the always-on AI application is embedded within the trusted execution environment (TEE) of the secure processor. This nesting allows the AI security function to leverage the existing TEE infrastructure for data protection, avoiding the need to design a completely separate security architecture and thus reducing overall device complexity while maintaining strong data protection.
Solution Approach 2:
The always-on AI application is designed to serve multiple functions including security protection, power management, and adaptability control within a unified architecture. By making the system multi-functional, the patent reduces the need for separate dedicated components for each function, thereby reducing device complexity while improving data protection through a consolidated secure processing approach.
Data Source
AI summary
Aspects of the present disclosure provide an apparatus that can execute an artificial intelligence (AI) model with IO changing. For example, the apparatus can include a first secured processor, a secured application embedded in the first secured processor and associated with an AI model, a secured memory configured to store an AI executable binary associated with the AI model, a second secured processor configured to execute the AI executable binary, a sub-system configured to trigger IO changing and trigger the second secured processor to execute the AI executable binary, IO meta data stored in the secured memory, an IO verifier configured to verify IO changing by determining the IO meta data, and an IO pre-fire module configured to patch the IO changing to the AI executable binary running on the second secured processor when the IO verifier determines that the IO changing matches the IO meta data.