Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

About start index #5

Closed
art-jang opened this issue Nov 1, 2021 · 2 comments
Closed

About start index #5

art-jang opened this issue Nov 1, 2021 · 2 comments

Comments

@art-jang
Copy link

art-jang commented Nov 1, 2021

Hi, thank you for your work.

I'm not sure whether the start index is False or True for KD loss calculation.

Could you let me know about it?

@ycmin95
Copy link
Collaborator

ycmin95 commented Nov 2, 2021

@JYJ-YeongJoon-Jang
We adopt the KD loss to distill the logits learnt by the BiLSTM layer to the convolution layer. Due to the spiky activations of CTC, we don't know whether blank labelled logits contain useful information, so we try distill with blank logits and without blank logits. Experimental results show that without blank logits can achieve better performance and we adopt this setting in this work.
There are some assumptions about this result: blank logits contain less information, or the unbanlanced ratio of blank and non-blank logits.

@art-jang
Copy link
Author

art-jang commented Nov 2, 2021 via email

@ycmin95 ycmin95 closed this as completed Nov 2, 2021
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants