Decentralized GAN Attribution Using Orthogonal Verification Keys
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing detection mechanisms for generative models are centralized, limiting scalability and accuracy in attributing machine-generated content to its source model, especially when models are distributed to end users.
Innovation Solution
A decentralized attribution system using a processor to compute orthogonal and data-compliant keys for generative adversarial networks (GANs), allowing users to verify outputs back to their source model through a public verification service.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If centralized detection mechanisms are used for generative models, then verification can be performed, but scalability is limited and attribution accuracy decreases when models are distributed to end users
Solution Approach 1:
The patent segments the centralized verification system into distributed components by assigning unique cryptographic keys to each end user's generative model. This allows each user to independently verify attributions locally without relying on a centralized server, thereby improving scalability while maintaining attribution accuracy through the decentralized key-based verification mechanism
Solution Approach 2:
The patent introduces cryptographic keys as an intermediary element that enables attribution verification. These keys serve as a mediator between the generative model and the verification process, allowing accurate attribution to be performed distributedly across multiple end users without requiring direct connection to a centralized detection mechanism
2Reliability
If orthogonal and data-compliant keys are computed for GANs, then correct attribution is ensured, but computational complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-computing orthogonal and data-compliant cryptographic keys during the model training phase. These keys are prepared in advance and embedded into the generative model architecture, so that during inference and verification, the attribution process can proceed efficiently without requiring complex real-time computations, thus maintaining attribution correctness while reducing operational computational complexity
Data Source
AI summary
A system and associated methods for decentralized attribution of GAN models is disclosed. Given a group of models derived from the same dataset and published by different users, attributability is achieved when a public verification service associated with each model (a linear classifier) returns positive only for outputs of that model. Each model is parameterized by keys distributed by a registry. The keys are computed from first-order sufficient conditions for decentralized attribution. The keys are orthogonal or opposite to each other and belong to a subspace dependent on the data distribution and the architecture of the generative model.