Sander Dieleman
sander.ai · Personal blog · on HN since 2023
Sander Dieleman, a machine learning researcher, writes detailed explainers on diffusion models, latent generative modelling, and related topics.
What HN saysHN commenters recommend it for alternative interpretations of diffusion and cite specific posts on latents, perspectives, and the spectral autoregression view.
27HN people shared it
9linked it in comments
25stories submitted
375story points
6 months agolast shared
2023people per year2026
Most-shared posts
- 1Generative Modelling in Latent Space9 people · 31 points · first shared 2025 · discussion on HN
- 2Diffusion Is Spectral Autoregression5 people · 225 points · first shared 2024 · 62 comments on HN
- 3Perspectives on Diffusion4 people · 2 points · first shared 2023 · discussion on HN
- 4Guidance: A cheat code for diffusion models3 people · 13 points · first shared 2023 · discussion on HN
- 5Diffusion Language Models3 people · 5 points · first shared 2023 · discussion on HN
- 6Generating music in the waveform domain (2020)2 people · 92 points · first shared 2024 · 41 comments on HN
- 7Musings on Typicality2 people · 3 points · first shared 2023 · discussion on HN
- 8The Paradox of Diffusion Distillation2 people · 1 points · first shared 2024 · discussion on HN
- 9Diffusion2 people · first shared 2023
- 10Noise schedules considered harmful (2024)1 people · 1 points · first shared 2026 · discussion on HN
- 11Learning the Integral of a Diffusion Model1 people · 1 points · first shared 2026 · discussion on HN
- 12The Geometry of Diffusion Guidance1 people · 1 points · first shared 2023 · discussion on HN
- 13Slides meetup1 people · first shared 2026
What HN people wrote when they linked it
vvolhejn · 2025-10-22
There's a great blog post from Sander Dieleman about exactly this - why do we need a two step pipeline, in particular for images and audio? https://sander.ai/2025/04/15/latents.html
For text, there are a few papers that train the tokenization and language model end-to-end, see: https://arxiv.org/abs/2305.07185
sl8r · 2025-02-15
Hey! I’m the post author. Highly recommend Sander Dieleman’s blog for alternative interpretations https://sander.ai/2023/07/20/perspectives.html
I personally find the SDEs the most intuitive, and the deterministic ODE / consistency models / rectified flow stuff as ideas that are easier to understand after the SDEs. But not everyone agrees!
vivzkestrel · 2026-05-07
- just a headsup
- your links to the slides for deeplearning you did here https://sander.ai/2014/05/29/slides-meetup.html are broken
benanne · 2026-05-06
I briefly covered that connection in an earlier blog post: https://sander.ai/2023/07/20/perspectives.html#flow
... but it's definitely something that might deserve a longer-form treatment at some point :)
tripplyons · 2025-10-20
Here's a blog post I liked that explains a connection: https://sander.ai/2024/09/02/spectral-autoregression.html
They call diffusion a form of "spectral autoregression", because it tends to first predict lower frequency features, and later predict higher frequency features.
throwaway2562 · 2025-03-12
I’m trying to understand what the ‘spectral’ interpretation of IMM is: but perhaps I shouldn’t
https://sander.ai/2024/09/02/spectral-autoregression.html
GaggiX · 2024-03-27
There are so many similar diffusion model distillation techniques these days that it's become hard to tell the difference between them, this is probably the fourth example I've seen that uses adversarial loss to distill the model, the others being UFOgen, some other work by StabilityAI on SDXL-turbo and similar, SDXL-Lightning by ByteDance. I found this blog page that explains some of the differences: https://sander.ai/2024/02/28/paradox.html
SleekEagle · 2023-04-19
https://sander.ai/2022/01/31/diffusion.html
The differences are thoroughly discussed here. From my point of view, while there is this surface-level similarity, the training process for diffusion models is directly a result of the mathematical framework on which they are defined which is distinct from denoising autoencoders.
I'd suggest reading the above blog if you are interested, it is thorough!