Questions about the paper and the code

[JavaZeroo]

Thank you very much for your great work! I'm just starting to learn diffusion bridge. there are a few things I don't understand that I hope you can answer:

1. why is the learn_score_marginal function used only in the repeat_ou.ipynb file? Based on my understanding of your paper, I think a complete training process should first use learn_score_transition and then learn_score_marginal, so that the diffusion bridge can be predicted completely.
2. what does the learn_full_score_transition function mean? I didn't see anything about it in the paper

---

Thank you very much for your valuable time! Looking forward to your reply!

[jeremyhengjm]

Hi there,

1. the function learn_score_marginal is only required when one wants a forward time approximation of the diffusion bridge. This was used only in the Ornstein-Uhlenbeck experiments as we found that (for our purposes) it is typically sufficient and better to approximate the backward time representation of the diffusion bridge. That said the forward time representation could be useful for other problems.
2. Yes please ignore the function learn_full_score_transition as it was used to experiment with an idea that we do not intend to have in the paper.

I hope you find these responses helpful. Thanks for your interest in our work.

[JavaZeroo]

Hi there,

1. the function learn_score_marginal is only required when one wants a forward time approximation of the diffusion bridge. This was used only in the Ornstein-Uhlenbeck experiments as we found that (for our purposes) it is typically sufficient and better to approximate the backward time representation of the diffusion bridge. That said the forward time representation could be useful for other problems.
2. Yes please ignore the function learn_full_score_transition as it was used to experiment with an idea that we do not intend to have in the paper.

I hope you find these responses helpful. Thanks for your interest in our work.

Thank you for your reply, it helps me a lot!