Faster Diffusion - presentation of the Denoising Diffusion Implicit Models paper

Here, we talk about Denoising Diffusion Implicit Models, a kind of diffusion models introduced by Song and al (2021)

This variation leads to a shorter sampling time compared to the original Denoising Diffusion Probabilistic Model (Ho and al, 2020), among other interesting properties.

[MonkkSoori]

This was an excellent overview and explanation of DDIM after I learned about DDPM. Thank you.

[talktovipin]

Nice explanation. Looking forward for similar explanation on Classifier-Free Diffusion Guidance.

[wonjun]

What a brilliant explanation. Thank you so much!!!

[clemacort]

Great explanation, thank you very much !

[vivekmittal]

Interesting talk, very well explained. Thank you :)

[charlherbst]

If you make the sampling deterministic, set eta=0, do you just generate the same training data? Does eta>0 to say you generated novel images?

[TheAero]

I may understand diffusion at high level, however the math just seem kinda random.