About hierarchy balancing

[hankyul]

Hello, Thank you for sharing your work!!!! 😀

During reading your paper, I got a question about some equation. (shown below picture)

My questions in equation (2) are follow:

• Does k means hierarchical layer number?
• If k is large, O_k means softmax at low higher hierarchies in ?

Again, thank you for sharing your great work!!!

[Tal]

Does k means hierarchical layer number - yes

The arxiv version of the paper had a mistake. i already corrected it in the NeurIPS version:
https://openreview.net/pdf?id=Zkj_VcZ6ol

Maybe i should also release a new arxiv version, to prevent further confusion.

kudos for the thorough reading.

[hankyul]

@mrT23 Thank you for your kind reply. I understand concept from your explanation.