socathie/circomlib-ml

Circom Circuits Library for Machine Learning

circomlib-ml's Issues

Floating number supports

Do these circuits supports floating numbers, or if i meets a floating number when calculating (such as dividing)? how should i deal with this problem

Conv1D doesn't work as expectation.

Source code location:
https://github.com/socathie/circomlib-ml/blob/master/circuits/Conv1D.circom

According to your test case: Conv1D(20, 3, 2, 4, 3), it seems in[19][*] are never used, according to the definition of Conv, I think it is an unexpected behavior.

Understanding Applications

Hi,

I'm curious what sort of applications this allows, and how it might be used in practice. I'm rather new to ZKP, and only understand the basics. I could be wrong, but it seems like you can verify that the output of a network (public statement) was computed for some unknown input and training data (witness) with a publically known ML model (public circuit). Is that right? Also what are some example applications.

I'm also new to Circom.

Lack of PReLU

Lack of PReLU.circom
reference from pytorch: https://pytorch.org/docs/stable/generated/torch.nn.PReLU.html