I am currently a master's student researching zkML.
I recently came across your project, DeepProve, and found it highly valuable for my work. I have two questions I would like to inquire about:
-
DeepProve focuses on proving the inference process of a model. Would it be possible to modify or extend it to provide proofs for the training process as well?
-
In the context of zero-knowledge proofs for machine learning, is it currently feasible to prove the security of the overall process by verifying only part of the data? For example, in model inference, could sampling a few instances be sufficient to ensure the integrity of the entire inference process?
Look forward to your response!
I am currently a master's student researching zkML.
I recently came across your project, DeepProve, and found it highly valuable for my work. I have two questions I would like to inquire about:
DeepProve focuses on proving the inference process of a model. Would it be possible to modify or extend it to provide proofs for the training process as well?
In the context of zero-knowledge proofs for machine learning, is it currently feasible to prove the security of the overall process by verifying only part of the data? For example, in model inference, could sampling a few instances be sufficient to ensure the integrity of the entire inference process?
Look forward to your response!