Iterative $α$-(de)Blending: a Minimalist Deterministic Diffusion Model · HackerPapers discussions

Iterative $α$-(de)Blending: a Minimalist Deterministic Diffusion Model — mentioned by stratos123

Sat, 10 Oct 2026 11:06:15 GMT

For more info on why flow-matching is more stable than DDIM, see Heitz 2023, which nicely explains how they're almost equivalent but DDIM corresponds to a differential equation with a 1/α term that diverges at the start of the denoising process, while flow-matching/IADB doesn't diverge: https://arxiv.org/abs/2305.03486

Paper and discussion history. A mention is not an endorsement.

Iterative $α$-(de)Blending: a Minimalist Deterministic Diffusion Model — mentioned by strangecasts

Mon, 11 Mar 2024 22:35:31 GMT

Super interesting! Immediately reminded of Iterative alpha-(de)Blending [1] which also sets out to set up a conceptually simpler diffusion model, and also arrives at formulating it as an approximate iterative projection process - I think this approach allows for more interesting experiments like the denoiser error analysis, though. [1] https://arxiv.org/pdf/2305.03486.pdf

Paper and discussion history. A mention is not an endorsement.