AI Model Watermark Verification for Data Processing Accelerators
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
There is a lack of effective digital rights protection for machine-learning models, and there is no proof that results produced by data processing accelerators are protected by a 'root of trust' system, making it difficult to verify the authenticity and authorization of AI models used in secondary processing systems.
Innovation Solution
A watermarking system is implemented where a watermark is embedded into AI models, and a signature is generated and verified to authenticate the model, ensuring that only authorized models are used in data processing accelerators, with the watermark being extracted and verified before inference tasks are performed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If machine-learning models are made portable and reusable without authorization, then ease of operation and adaptability improve, but digital rights protection and reliability deteriorate
Solution Approach 1:
A watermark is embedded into the AI model before it is deployed or transferred to secondary processing systems. This preliminary action ensures that the model carries authentication information that can be verified later, allowing the model to be portable while maintaining digital rights protection through pre-established trust mechanisms
Solution Approach 2:
The watermark acts as an intermediary element between the AI model and the verification system. It provides a mechanism for secondary processing systems to verify the authenticity and authorization of the model without requiring direct authorization protocols, thus enabling portability while maintaining reliability through the intermediary verification layer
2Adaptability or versatility
If AI models are deployed in secondary processing systems without verification, then adaptability and ease of operation improve, but reliability and trustworthiness deteriorate
Solution Approach 1:
The watermark is embedded into the AI model in advance, before deployment to secondary processing systems. This allows the model to be adapted and deployed flexibly across different systems while the pre-embedded watermark enables subsequent verification of authenticity and authorization, maintaining reliability without restricting adaptability
Solution Approach 2:
The watermark serves as a verifiable copy or representation of the model's authorization credentials. Secondary processing systems can verify the watermark to confirm the model's authenticity without needing to access the original authorization source, enabling flexible deployment while maintaining trust through verifiable copies
Data Source
AI summary
Embodiments of the disclosure relate to verifying a watermark of an artificial intelligence (AI) model for a data processing (DP) accelerator. In one embodiment, a system receives an inference request from an application. The system extracts the watermark from an AI model having the watermark. The system verifies the extracted watermark based on a policy. The system applies the AI model having a watermark to a set of inference inputs to generate inference results. The system sends a verification proof and the inference results to the application.