seonwoo-min/GVRT

Inquiry about pretraining for Discriminator in sentence classification

Closed this issue · 2 comments

Hello,

First of all, thank you for the fantastic article!
I wanted to inquire about whether the Discriminator needs pretraining on sentence classification tasks. I noticed in the article "Generating visual explanations" from 2016, there was mention of pretraining. Could you please clarify whether pretraining for Discriminator is necessary, and if so, what approach would be recommended?
Thank you for your attention to this matter.

Hey, @JackeyWang777. Thanks for your interest in our work!
In our case, we do not pre-train the discriminator, a.k.a., sentence classifier.
As you can check in our codes, all the additional modules (i.e., two classifiers, two projectors, an explainer, and a discriminator) are trained jointly from scratch without loading any pretrained weights.

Thanks,
Seonwoo

Hey, @JackeyWang777. Thanks for your interest in our work! In our case, we do not pre-train the discriminator, a.k.a., sentence classifier. As you can check in our codes, all the additional modules (i.e., two classifiers, two projectors, an explainer, and a discriminator) are trained jointly from scratch without loading any pretrained weights.

Thanks, Seonwoo

정말 감사합니다