Changmin Cho

While the ZK community and ML community are both growing rapidly, there are few bridges connecting the two disciplines. One of them is ZK-ML, which validates correctness of ML by ZK. On ZK-ML, ZK can make sure whether data is actually used on running ML model. But the problem is applying ZK on ML requires some stiff backgrounds and principles if developed from scratch. Our goal is to minimalize the gap between existing ML technology and newly applied ZK technology: Therefore, we made a tool 'torch2circom', which bridges the gap. Developers can run torch2circom on their own ML model, as they acquires the proof generator written on circom. This can be utilized in smart contracts: When user submits an input, ML developer can give the result and the proof that they have ran the model. When the proof have been validated, the smart contract can be made so that users send money according to the validation. Therefore, torch2circom is not only useful to ML developers who wants to prove their run to the users but also to the users to wants to ensure that their data has actually gone through a feasible neural networks.