Blog: https://sander.ai/ 🐦: https://x.com/sedielem Research Scientist at Google DeepMind (WaveNet, Imagen 3, Veo, ...). I tweet about deep learning (research + software), music, generative models (personal account).https://bsky.app/profile/sedielem.bsky.socialhttps://bsky.app/profile/sedielem.bsky.social/post/3ml6fiotfo225My first blog post in over a year is a deep dive on flow maps🗺️, or how to learn the integral of a diffusion model to enable faster sampling and several other cool tricks. It's the longest one yet👀 Let me know what you think! https://sander.ai/2026/05/06/flow-maps.html06 May 2026 09:34 +0000at://did:plc:tl4gcezwgrweh4ft7eng7omj/app.bsky.feed.post/3ml6fiotfo225https://bsky.app/profile/sedielem.bsky.social/post/3mh7g2dv4nc25In October, I gave a talk at ML in PL in Warsaw: a whirlwind tour of what goes into training image and video generation models at scale. 📺 video: https://www.youtube.com/watch?v=qFIT3mSwWEk 🖼️ slides: https://docs.google.com/presentation/d/1hg0Awaia0C_uQsX00aSUH89SvAGXVCadFR83pCGWO3U/edit?usp=sharing16 Mar 2026 21:35 +0000at://did:plc:tl4gcezwgrweh4ft7eng7omj/app.bsky.feed.post/3mh7g2dv4nc25https://bsky.app/profile/sedielem.bsky.social/post/3luztqoaj3s2hGreat blog post on rotary position embeddings (RoPE) in more than one dimension, with interactive visualisations, a bunch of experimental results, and code! https://jerryxio.ng/posts/nd-rope/28 Jul 2025 14:51 +0000at://did:plc:tl4gcezwgrweh4ft7eng7omj/app.bsky.feed.post/3luztqoaj3s2hhttps://bsky.app/profile/sedielem.bsky.social/post/3luvkpfvx5225I blog and give talks to help build people's intuition for diffusion models. YouTubers like @3blue1brown.com and Welch Labs have been a huge inspiration: their ability to make complex ideas in maths and physics approachable is unmatched. Really great to see them tackle this topic! [contains quote post or other embedded content]26 Jul 2025 21:59 +0000at://did:plc:tl4gcezwgrweh4ft7eng7omj/app.bsky.feed.post/3luvkpfvx5225https://bsky.app/profile/sedielem.bsky.social/post/3ltzu5z5txk2nHello #ICML2025👋, anyone up for a diffusion circle? We'll just sit down somewhere and talk shop. 🕒Join us at 3PM on Thursday July 17. We'll meet here (see photo, near the west building's west entrance), and venture out from there to find a good spot to sit. Tell your friends!15 Jul 2025 21:33 +0000at://did:plc:tl4gcezwgrweh4ft7eng7omj/app.bsky.feed.post/3ltzu5z5txk2nhttps://bsky.app/profile/sedielem.bsky.social/post/3lta4wz5pas2qDiffusion models have analytical solutions, but they involve sums over the entire training set, and they don't generalise at all. They are mainly useful to help us understand how practical diffusion models generalise. Nice blog + code by Raymond Fan: https://rfangit.github.io/blog/2025/optimal_flow_matching/05 Jul 2025 16:01 +0000at://did:plc:tl4gcezwgrweh4ft7eng7omj/app.bsky.feed.post/3lta4wz5pas2qhttps://bsky.app/profile/sedielem.bsky.social/post/3lp5f7v7msk2uHere's the third and final part of Slater Stich's "History of diffusion" interview series! The other two interviewees' research played a pivotal role in the rise of diffusion models, whereas I just like to yap about them 😬 this was a wonderful opportunity to do exactly that! https://www.youtube.com/watch?v=ktPGNhe11cQ14 May 2025 16:11 +0000at://did:plc:tl4gcezwgrweh4ft7eng7omj/app.bsky.feed.post/3lp5f7v7msk2uhttps://bsky.app/profile/sedielem.bsky.social/post/3lp4y4la2as24The ML for audio 🗣️🎵🔊 workshop is back at ICML 2025 in Vancouver! It will take place on Saturday, July 19. Featuring invited talks from Dan Ellis, Albert Gu, James Betker, Laura Laurenti and Pratyusha Sharma. Submission deadline: May 23 (Friday next week) mlforaudioworkshop.github.io https://mlforaudioworkshop.github.io/14 May 2025 12:16 +0000at://did:plc:tl4gcezwgrweh4ft7eng7omj/app.bsky.feed.post/3lp4y4la2as24https://bsky.app/profile/sedielem.bsky.social/post/3lnnbuurscs2pOne weird trick for better diffusion models: concatenate some DINOv2 features to your latent channels! Combining latents with PCA components extracted from DINOv2 features yields faster training and better samples. Also enables a new guidance strategy. Simple and effective! [contains quote post or other embedded content]25 Apr 2025 13:03 +0000at://did:plc:tl4gcezwgrweh4ft7eng7omj/app.bsky.feed.post/3lnnbuurscs2phttps://bsky.app/profile/sedielem.bsky.social/post/3lmts2lds2c22New blog post: let's talk about latents! https://sander.ai/2025/04/15/latents.html15 Apr 2025 09:43 +0000at://did:plc:tl4gcezwgrweh4ft7eng7omj/app.bsky.feed.post/3lmts2lds2c22https://bsky.app/profile/sedielem.bsky.social/post/3lmrzay3r4227Amazing interview with Yang Song, one of the key researchers we have to thank for diffusion models. The most important lesson: be fearless! The community's view on score matching was quite pessimistic at the time, he went against the grain and made it work at scale! https://www.youtube.com/watch?v=ud6z5SkjoZI14 Apr 2025 16:47 +0000at://did:plc:tl4gcezwgrweh4ft7eng7omj/app.bsky.feed.post/3lmrzay3r4227https://bsky.app/profile/sedielem.bsky.social/post/3lipii7hyis2cWe are hiring on the Generative Media team in London: https://boards.greenhouse.io/deepmind/jobs/6192747 We work on Imagen, Veo, Lyria and all that good stuff. Come work with us! If you're interested, apply before Feb 28.21 Feb 2025 19:00 +0000at://did:plc:tl4gcezwgrweh4ft7eng7omj/app.bsky.feed.post/3lipii7hyis2chttps://bsky.app/profile/sedielem.bsky.social/post/3lhu72iiffs2yGreat interview with @jascha.sohldickstein.com about diffusion models! This is the first in a series: similar interviews with Yang Song and yours truly will follow soon. (One of these is not like the others -- both of them basically invented the field, and I occasionally write a blog post 🥲) https://www.youtube.com/watch?v=VpYNlHIHT7o10 Feb 2025 22:30 +0000at://did:plc:tl4gcezwgrweh4ft7eng7omj/app.bsky.feed.post/3lhu72iiffs2yhttps://bsky.app/profile/sedielem.bsky.social/post/3lgebig5xk226📢PSA: #NeurIPS2024 recordings are now publicly available! The workshops always have tons of interesting things on at once, so the FOMO is real😵‍💫 Luckily it's all recorded, so I've been catching up on what I missed. Thread below with some personal highlights🧵22 Jan 2025 21:06 +0000at://did:plc:tl4gcezwgrweh4ft7eng7omj/app.bsky.feed.post/3lgebig5xk226https://bsky.app/profile/sedielem.bsky.social/post/3leokondi522hWhy do diffusion models generalise at all? It's not obvious that they would. It turns out underfitting plays an important role, as well as the architectural inductive biases of locality and translation equivariance. What other kinds of symmetry and structure could we hardcode? 🤔 [contains quote post or other embedded content]01 Jan 2025 12:26 +0000at://did:plc:tl4gcezwgrweh4ft7eng7omj/app.bsky.feed.post/3leokondi522hhttps://bsky.app/profile/sedielem.bsky.social/post/3ldkjxw7rxs2uThe recording of my #NeurIPS2024 workshop talk on multimodal iterative refinement is now available to everyone who registered: https://neurips.cc/virtual/2024/workshop/84747 My talk starts at 1:10:45 into the recording. I believe this will be made publicly available eventually, but I'm not sure when exactly! [contains quote post or other embedded content]18 Dec 2024 04:38 +0000at://did:plc:tl4gcezwgrweh4ft7eng7omj/app.bsky.feed.post/3ldkjxw7rxs2uhttps://bsky.app/profile/sedielem.bsky.social/post/3ldguhk7qak26Here's Veo 2, the latest version of our video generation model, as well as a substantial upgrade for Imagen 3 🧑‍🍳🚢 (Did I mention we are hiring on the Generative Media team, btw 👀) https://blog.google/technology/google-labs/video-image-generation-update-december-2024/16 Dec 2024 17:35 +0000at://did:plc:tl4gcezwgrweh4ft7eng7omj/app.bsky.feed.post/3ldguhk7qak26https://bsky.app/profile/sedielem.bsky.social/post/3ldag4cqw5k26I've been getting a lot of questions about autoregression vs diffusion at #NeurIPS2024 this week! I'm speaking at the adaptive foundation models workshop at 9AM tomorrow (West Hall A), about what happens when we combine modalities and modelling paradigms. adaptive-foundation-models.org https://adaptive-foundation-models.org14 Dec 2024 04:02 +0000at://did:plc:tl4gcezwgrweh4ft7eng7omj/app.bsky.feed.post/3ldag4cqw5k26https://bsky.app/profile/sedielem.bsky.social/post/3ld7fzspknc2bIf you're at #NeurIPS2024, join us this afternoon to talk diffusion, flows and all that jazz. Be there or be non-circular! [contains quote post or other embedded content]13 Dec 2024 18:28 +0000at://did:plc:tl4gcezwgrweh4ft7eng7omj/app.bsky.feed.post/3ld7fzspknc2bhttps://bsky.app/profile/sedielem.bsky.social/post/3ld33u6ijgs2rWhen a bunch of diffusers sit down and talk shop, their flow cannot be matched😎 It's time for the #NeurIPS2024 diffusion circle! 🕒Join us at 3PM on Friday December 13. We'll meet near this thing, and venture out from there and find a good spot to sit. Tell your friends!12 Dec 2024 01:15 +0000at://did:plc:tl4gcezwgrweh4ft7eng7omj/app.bsky.feed.post/3ld33u6ijgs2rhttps://bsky.app/profile/sedielem.bsky.social/post/3lcdvlaxyp22yBetter VQ-VAEs with this one weird rotation trick! I missed this when it came out, but I love papers like this: a simple change to an already powerful technique, that significantly improves results without introducing complexity or hyperparameters. [contains quote post or other embedded content]02 Dec 2024 19:52 +0000at://did:plc:tl4gcezwgrweh4ft7eng7omj/app.bsky.feed.post/3lcdvlaxyp22yhttps://bsky.app/profile/sedielem.bsky.social/post/3lcdtdeui722fThere is a lot of great writing on flow matching out there all of a sudden! This post clarifies the connection with diffusion models -- they are essentially two different ways to describe the same class of models. [contains quote post or other embedded content]02 Dec 2024 19:11 +0000at://did:plc:tl4gcezwgrweh4ft7eng7omj/app.bsky.feed.post/3lcdtdeui722fhttps://bsky.app/profile/sedielem.bsky.social/post/3lcdrdnkn7s2iIn arxiv.org/abs/2303.00848, @dpkingma.bsky.social and @ruiqigao.bsky.social had suggested that noise augmentation could be used to make other likelihood-based models optimise perceptually weighted losses, like diffusion models do. So cool to see this working well in practice! [contains quote post or other embedded content]02 Dec 2024 18:36 +0000at://did:plc:tl4gcezwgrweh4ft7eng7omj/app.bsky.feed.post/3lcdrdnkn7s2ihttps://bsky.app/profile/sedielem.bsky.social/post/3lcbhlz2k6s2yOne of these is not like the others 😁 [contains quote post or other embedded content]01 Dec 2024 20:36 +0000at://did:plc:tl4gcezwgrweh4ft7eng7omj/app.bsky.feed.post/3lcbhlz2k6s2y