AI Model Control Through Microcode Translation and License Protection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge of protecting AI models implemented on client devices from being copied or stolen is addressed by providing control of input, output, and processing through a system that translates AI models into intermediate commands and microcode executable by hardware devices, while enforcing security requirements and licenses.
Innovation Solution
AI models are translated into intermediate commands and microcode, with hardware devices executing these commands to generate outputs, and security is enforced through licenses specifying protection levels, device requirements, and hardware-based restrictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If AI models are deployed on client devices, then accessibility and functionality are improved, but security and protection against copying are worsened
Solution Approach 1:
The AI model is divided into multiple protected portions (first portion, second portion, etc.) with different protection levels. Each portion can be independently protected and controlled, allowing selective access while maintaining overall model functionality. This segmentation enables the system to provide model accessibility to authorized users while preventing unauthorized copying of specific critical portions.
Solution Approach 2:
A translation system acts as an intermediary between the AI model and the hardware device. The model is translated to intermediate language commands, then to hardware-specific commands, and finally to microcode. This intermediary translation layer ensures that the original model remains protected while enabling execution on client devices, preventing direct access to the model's core components.
2Object-affected harmful factors
If multiple protection levels are applied to AI model portions, then security is improved, but system complexity is worsened
Solution Approach 1:
Different portions of the AI model are assigned different protection levels based on their sensitivity and importance. Critical portions receive higher protection levels while less sensitive portions have lower protection. This local differentiation allows the system to maintain strong security where needed while reducing complexity in less critical areas, avoiding the need to protect the entire model at maximum security level.
3Object-affected harmful factors
If AI models are translated through multiple command sets, then protection is improved, but processing time is worsened
Solution Approach 1:
The translation of the AI model to intermediate language commands and hardware-specific commands is performed in advance before execution. This preliminary translation action, combined with the use of microcode for efficient hardware execution, reduces the time required during actual model execution. The protective translation layers are established once and reused across multiple executions.
Data Source
AI summary
Examples of the present disclosure describe systems and methods for providing control of input, output, and processing of an AI model. In examples, a request to execute an AI model implemented by a client device is received, where the AI model is associated with one or more licenses that specify a protection level that is applied to one or more portions of the AI model during the AI model runtime. In response to the request, the AI model is translated to a first set of commands in an intermediate language. The first set of commands is translated into a second set of commands for a hardware device of the client device. The second set of commands is translated into microcode that is executable by the hardware device. The hardware device then executes the microcode to generate an output in furtherance of the request.