Is reconstruction loss equivalent to noise prediction loss?

[justin4ai]

Hello friends, first of all thanks for providing such a great project for the world.

Reading your paper, one question came to my mind - there is a reconstruction loss in your architecture figure but not in your training loss explanation where only noise prediction loss exists.

At the first glance I thought the ,,reconstruction loss'' literally means L2 loss or something so would anyone explain this?

Thanks, Junyeong Ahn