MLPBinaryConnect is full precision but not binary

[snownus]

seems MLPBinaryConnect is full precision but not binary?

[mengxiangming]

seems MLPBinaryConnect is full precision but not binary?

Different from STE where we have to create a special "binary" network architecture, we can simply create the network structure as usual (for example, MLPBinaryConnect). After that, if you want to obtain a binary neural network, you simply need to train the neural network using our BayesBiNN optimizer rather than the common continuous one like Adam. In this way, BayesBiNN is more flexible and convenient than the previous binary optimizers since it is model-agnostic.

[snownus]

I see. Thanks very much for the instant reply.

[snownus]

@mengxiangming could I know the python/pytorch enviroment of running the repo?

[mengxiangming]

@mengxiangming could I know the python/pytorch enviroment of running the repo?

Well, I think it can be run on normal settings and I forget the exact minimal requirement. I just tried running the repo on Pytorch '1.12.0' and Python 3.8.8 and it works.