Compare with model EPM？ about sigma HOT 3 CLOSED

Hi, thanks for your interest.

Maybe, I think that a reasonable and easy solution is to share the first stage model with EPM, i.e., the GA model.

from sigma.

@wymanCV
May I ask about the comparative experiment with model EPM in your paper? Is the model used in the original EPM paper or modified? Thank you. If it is modified, could you please provide the modified EPM?

from sigma.

Hi, as claimed in the implementation details, the pre-trained stage in our previous experiments removes BGM by disabling the following two items.

WITH_DOMAIN_INTERACTION False
MATCHING_CFG 'none'

We also tried to use EPM's 1st stage model, finding that using it can achieve similar final results.

from sigma.