Consensus: Validators evaluate these "masks" and vote on the best options.
Iterative Cycle: An intermediate layer of the neural network is used to randomly verify the work of validators (specifically, to check whether their updates are based on valid data).
Important clarification: While ZK proofs in this context verify the correctness of a specific computation, they do not prevent the submission of falsified data; the fourth mechanism (the iterative cycle) is designed to address this.
This is an interesting area of research aimed at enhancing participant privacy during AI model training. Naturally, there are some drawbacks—such as the lack of complete privacy due to the use of intermediate neural network layers, or the issue of data reuse.
The post this quotes
Terminus needs JavaScript to show replies and pictures.