How to transport or fuse between two biases?

[BossunWang]

Hi expert,
Nice work and novel view to use OT method.
As you mentioned on your paper "The bias of a neuron is set to zero in all of the experiments. It is possible to handle
it as a regular weight by keeping the corresponding input as 1" .
Do you have any new version to handle fusion of bias over this years?
Or How to do a regular weight by keeping the corresponding input as 1?

[sidak]

Hi,

Thanks for your interest and kind words. I have been wanting to support this, but haven't found time. One option is to simply apply the permutations maps obtained by aligning the weight matrices for a layer and use that to align the bias parameters as well (hit the bias vector by the corresponding permutation matrix).

However, it would also be nice to develop a version that makes use of the bias parameters while finding the alignment itself. Feel free to share your experience, perhaps with a patch, handling this :)

(I also foresee getting back to this sometime very soon, so I will push my update too)

Best,
Sidak