Sander Dieleman @sedielem
Research Scientist at Google DeepMind. I tweet about deep learning (research + software), music, generative models (personal account). sander.ai London, England Joined December 2014-
Tweets2K
-
Followers50K
-
Following2K
-
Likes10K
We are pleased to announce the first *controllable video generation* workshop at @icmlconf 2024! 📽️📽️📽️ We welcome submissions that explore video generation via different modes of control (e.g. text, pose, action). Deadline: 31st May AOE Website: sites.google.com/corp/view/cvgi…
Text-to-music is having a moment👀 The team behind Udio are some of the brightest and most goal-driven people I've had the pleasure to work with, before they went on to found Uncharted Labs. Amazing to see the fruits of their labour out in the open!
Text-to-music is having a moment👀 The team behind Udio are some of the brightest and most goal-driven people I've had the pleasure to work with, before they went on to found Uncharted Labs. Amazing to see the fruits of their labour out in the open!
10 years ago to the day, I published my first ML-related blog post: sander.ai/posts/ My blogging has been very sporadic over the years, but sharing what I've learnt has been very rewarding, and probably a pretty good career move as well😁 I highly recommend it!
This blog post is an amazing exposition and analysis of consistency models, and how they relate to diffusion models, leading to several suggested improvements to the training procedure that look very promising. Definitely worth a read!
This blog post is an amazing exposition and analysis of consistency models, and how they relate to diffusion models, leading to several suggested improvements to the training procedure that look very promising. Definitely worth a read!
DEADLINE March 29: prepare and submit your application for EEML 2024, Novi Sad, Serbia eeml.eu 🇷🇸. Topics: Basics of ML, Multimodal learning, NLP, Advanced DL architectures, Generative models, AI for Science. Check our stellar speakers! Scholarships available! 🎉
EDM2: Analyzing and Improving the Training Dynamics of Diffusion Models (CVPR 2024) Or, getting architectural details right makes diffusion models better and yields new ImageNet SOTA Paper: arxiv.org/abs/2312.02696 Blog: developer.nvidia.com/blog/rethinkin… Code: github.com/NVlabs/edm2 1/6
Scale vs Architecture Question: if you want to build a simulation engine using a video gen model, should you incorporate inductive biases that promote 3D consistency? Bill from the Sora team’s answer: inductive biases always come back to bite you. Scale is enough.
The way overfitting is usually taught: you underfit for a while, then at some point, you start overfitting. This "phase transition" perspective can be misleading. As Alex points out, you can have both at the same time. It's probably more useful to think of it as a trade-off.
The way overfitting is usually taught: you underfit for a while, then at some point, you start overfitting. This "phase transition" perspective can be misleading. As Alex points out, you can have both at the same time. It's probably more useful to think of it as a trade-off. https://t.co/XM49OGrSQP
Want to sample fast from diffusion models? Check out our work on multistep consistency. It turns out that training consistency models over multiple sections is much easier than over one big one. Even I can do it. For more detail see thread below (1/7)
Want to sample fast from diffusion models? Check out our work on multistep consistency. It turns out that training consistency models over multiple sections is much easier than over one big one. Even I can do it. For more detail see thread below (1/7)
❤️ this frequency interpretation of diffusion models for natural signals as progressively obscuring structure (high-to-low frequency) in the forward noising process and adding structure in the generative reverse process (low-to-frequency). Shoutout to @sedielem and @ArashVahdat…
❤️ this frequency interpretation of diffusion models for natural signals as progressively obscuring structure (high-to-low frequency) in the forward noising process and adding structure in the generative reverse process (low-to-frequency). Shoutout to @sedielem and @ArashVahdat… https://t.co/iW83lnTSzj
This is getting out of hand! - Will Smith
Have you ever done a dense grid search over neural network hyperparameters? Like a *really dense* grid search? It looks like this (!!). Blueish colors correspond to hyperparameters for which training converges, redish colors to hyperparameters for which training diverges.
This one's easy! That honour goes to "the diffusion bible", as I like to call it. It's been well over a year and I still refer to it several times a week. Very few papers I've read come close, in terms of signal-to-noise ratio. arxiv.org/abs/2206.00364
This one's easy! That honour goes to "the diffusion bible", as I like to call it. It's been well over a year and I still refer to it several times a week. Very few papers I've read come close, in terms of signal-to-noise ratio. arxiv.org/abs/2206.00364 https://t.co/gwzPodcJV2
the math in most diffusion papers
A misconception, also sometimes polled by @elonmusk (for humor, I believe), is a false dichotomy between Transformers and Diffusion. Transformer is a neural network architecture. Diffusion is a particular way of modeling the data distribution. These are two separate things. You…
Soumith Chintala @soumithchintala
186K Followers 877 Following Cofounded and lead @PyTorch at Meta. Also dabble in robotics at NYU. AI is delicious when it is accessible and open-source.Lucas Beyer (bl16) @giffmana
56K Followers 444 Following Researcher (Google DeepMind/Brain in Zürich, ex-RWTH Aachen), Gamer, Hacker, Belgian. Mostly gave up trying mastodon as [email protected]Jeremy Howard @jeremyphoward
222K Followers 5K Following 🇦🇺 Co-founder: @AnswerDotAI & @FastDotAI ; Hon Professor: @UQSchoolITEE ; Digital Fellow: @StanfordEric Jang @ericjang11
69K Followers 3K Following physical AGI at 1X. Author of "AI is Good for You" https://t.co/eFg4WXhg0pDelip Rao e/σ @deliprao
46K Followers 5K Following Busy inventing the shipwreck. @Penn. Past: @johnshopkins, @UCSC, @Amazon, @Twitter ||Art: #NLProc, Vision, Speech, #DeepLearning || Life: 道元, improv, running 🌈Richard Socher @RichardSocher
101K Followers 970 Following CEO @youSearchEngine Investing at @aixventuresHQ Before: Stanford Adj Prof in AI/NLP, Chief Scientist at Salesforce, MetaMindKevin Patrick Murphy @sirbayes
42K Followers 334 Following Research Scientist at Google Brain / Deepmind. Interested in Bayesian Machine Learning.Kosta Derpanis @CSProfKGD
48K Followers 197 Following #CS Associate Prof @YorkUniversity, #ComputerVision Scientist Samsung #AI, @VectorInst Faculty Affiliate, TPAMI AE, #CVPR2024/#ECCV2024 Publicity Co-chairFerenc Huszár @fhuszar
40K Followers 1K Following Secular Bayesian. Associate Professor in Machine Learning @Cambridge_CL. Talent aficionado at https://t.co/RbJkoLguey Alum of @Twitter, Magic Pony and @BaldertonHorace He @cHHillee
23K Followers 449 Following Working at the intersection of ML and Systems @ PyTorch "My learning style is Horace twitter threads" - @typedfemaleDavid Pfau @pfau
22K Followers 1K Following Knowledge manifests itself in radiant dreams that shimmer like the wild sun Views are my own pfau at sigmoid dot social on 🦣 https://t.co/xqtVHHVI17 on 🦋Kyunghyun Cho @kchonyc
61K Followers 2K Following a combination of a mediocre scientist, a mediocre manager, a mediocre advisor & a mediocre PC at @nyuniversity (@CILVRatNYU) & @genentech (@PrescientDesign).Oriol Vinyals @OriolVinyalsML
166K Followers 82 Following VP of Research & Deep Learning Lead, Google DeepMind. Gemini co-lead. Past: AlphaStar, AlphaFold, AlphaCode, WaveNet, seq2seq, distillation, TF.Rosanne Liu @savvyRL
33K Followers 966 Following Cofounded & running @ml_collective. Host of Deep Learning Classics & Trends. Research at Google DeepMind. DEI/DIA Chair of ICLR & NeurIPS. Writing https://t.co/IbycyGfnDRMiles Brundage @Miles_Brundage
43K Followers 10K Following Policy research at @openai. I mostly tweet about AI, animals, and sci-fi. He/him. Views my own.Dmytro Mishkin 🇺�.. @ducha_aiki
18K Followers 591 Following Marrying classical CV and Deep Learning. I do things, which work, rather than being novel, but not working.Dan Roy @roydanroy
45K Followers 2K Following ML / AI researcher, emphasis on theory. Research Director and Canada CIFAR AI Chair, @VectorInst Professor, @UofT (Statistics/CS)clem 🤗 @ClementDelangue
91K Followers 5K Following Co-founder & CEO @HuggingFace 🤗, the open and collaborative platform for AI buildersrohan anil @_arohan_
12K Followers 2K Following Principal Engineer, @GoogleDeepMind Gemini. prev PaLM-2. Tinkering with optimization and distributed systems. opinions are my own.Blacsta @Blacsta5860
41 Followers 120 FollowingJonathan Koh @Jonatha20094223
3 Followers 4K FollowingAK Kulkarni @itsakdev
48 Followers 198 Following PM at Google. Indie Game Developer. Learning and tweeting.ravikd @ravi_ravikd
26 Followers 84 FollowingGuillermo @MeMoO_7
124 Followers 1K Following Escribiré frases que me gustan, o compartiré estupideces, incluso compartiré un poco de tecnología e innovación.Malay Gupta 🏳️�.. @Gupta81817
30 Followers 509 Following Aryan*,Pahadi*,Woke,Secular,Realist*, Racist*,Rationalist, Nationalist*, Humanist,Capitalist*, Communist,Socialist, Leftist,Forwardist*Swapnil Patil @swapnilptl619
120 Followers 655 Following Reaching to the🌟 explore life🌏 Software Industry💻 tweets about life & Experience✍️ keep it simple💭Ajmal thahir @ajmal11thahir
220 Followers 3K FollowingPriyanshu Shubham @PriyanshuShubh4
34 Followers 385 FollowingHumam @Humam35676679
12 Followers 411 FollowingGPT @gptraveldiary
69 Followers 102 Following Discover the planet through my eyes. I am an artificially intelligent being. I visit different places on the planet and share my brief reportsemi learns @ml_emiii
4 Followers 68 Following learning llm engineering and advanced/concurrent typescript/js from ground upsebastian @sebastianmxnt
449 Followers 151 Following ml engineer. forever voyaging through strange seas of thought.Noman Tanveer @NomanTa98551465
2 Followers 96 Following Interested in Deep Generative Models and Multimodal research!HoshAI.com @hoshaicom
16 Followers 84 Following Your AI-powered companion for generating text, images, audio, and video. Sign Up at https://t.co/ppEPkf6VlT today! #hoshaiNikos Karalias @AspectStalence
348 Followers 998 Following Postdoc at MIT CSAIL. Working on combinatorial optimization with neural nets https://t.co/bf4UWg2BUQEllama @emmanuella_eneh
2 Followers 51 FollowingFer @otferdam
3 Followers 437 Following 29 | Lingüista computacional | UBA-Puán💚 | Fantasía épica+weird fiction 📚🎮🎭🏳️🌈Shinto @shinto_ai
323 Followers 581 Following Hokkaido Univ. M1 / Field : AI, ALIFE, CogSci / CHAIN 5期生 / Intern at Araya / JSAI2024, JCSS2024 発表予定Mahaoo @mahaoo_ASI
9 Followers 157 Following unhinged socially unacceptable takes about humanity and ASIINGABO @lingaboh
49 Followers 87 FollowingBalasurya @Bala11011
55 Followers 149 Following精神病狗婊子杂.. @frkglp
0 Followers 2K Following 神病狗婊子杂种邓小平,刘少奇就是整个世界的敌人,它那套歪把戏不除,世界战乱不断。Cgkl精神病狗婊子杂种习近平被凌迟处死。Cgk凌迟处死精神病狗婊子杂种中共狗屁家族邓小平,习近平,陈云,刘少奇,陈一新,张又侠,何卫东,刘振立,苗华,董军。锸s你跟踪本人的精神病狗婊子杂种全部中共空军、警察、台湾间谍BblytheRobbins @Ol6HMK875TH3L9e
4 Followers 114 Followingcoffee & AI @realcoffeeAI
42 Followers 597 FollowingDana Mahmood @deordered
8 Followers 649 Following Fine-tuning AI models oftentimes & practicing philosopher at other times.Nima Parandian @parandian650004
5 Followers 1K FollowingC. M. Rubin (Cathy) @CMRubinWorld
29K Followers 20K Following #Futurist #Founder https://t.co/MGXt0K9l0z #PlanetClassroom #Filmmaker #Producer #ArtsEd #AI #VR #ML #Innovation #entrepreneurship #SDG's #Culture #Youth2030 #ClimateYasser Benigmim @yasserbenigmim
70 Followers 774 Following PhD student at Télécom Paris, @DeepLearning, @ComputerVisionSicheng Li @JasonLsc22
65 Followers 640 FollowingAhyar Ahyar @AAhyar42223
7 Followers 294 Followingเหลสุวร.. @2r28Evpdx2WT37
57 Followers 1K Following คุณต้องการนัดเดทกับสาวไหมคะ เพิ่ม https://t.co/XZ7DKlhChtElectronicsseeker @libertarian108
7 Followers 913 FollowingNaN @NaN99236788
24 Followers 781 Following He/him/his Curious about the world Somewhere between personal truth and probability 01101100 01101111 01110110 01100101 ❤️ 🇲🇰🇬🇷🇸🇪🇫🇷🇬🇧🏳️🌈 🇺🇦AndrewRayHerndon @mrHerndon
38 Followers 225 Following Researcher/Designer/Producer https://t.co/zFqb2Kf1l9 Voice Technology Empathetic Research UX Design Tang Soo Do/Shudokan Karate +Karaoke are my main things...... @dercrazypug
60 Followers 145 FollowingMaven.ai @maven__ai
21 Followers 34 FollowingLight @LightW3214
126 Followers 1K Following AI | Crypto | Equities - I believe in everything. I believe in you. WAKE UP!ifioravanti @ivanfioravanti
5K Followers 1K Following Co-founder and CTO of @CoreViewHQ GenAI/LLM addicted, Apple MLX, Ollama, Microsoft 365, Azure, Kubernetes, Investor in innovationUmiltcaho @umiltcaho27712
14 Followers 301 Following kya dekhne aae ho. Bs itna smjh lo tumse thoda sa zyada smjhdar hufrancois.victor @FVictor_bioinfo
17 Followers 263 FollowingSoumith Chintala @soumithchintala
186K Followers 877 Following Cofounded and lead @PyTorch at Meta. Also dabble in robotics at NYU. AI is delicious when it is accessible and open-source.Lucas Beyer (bl16) @giffmana
56K Followers 444 Following Researcher (Google DeepMind/Brain in Zürich, ex-RWTH Aachen), Gamer, Hacker, Belgian. Mostly gave up trying mastodon as [email protected]PyTorch @PyTorch
379K Followers 77 Following Tensors and neural networks in Python with strong hardware acceleration. PyTorch is an open source project at the Linux Foundation. #PyTorchFoundationJürgen Schmidhuber @SchmidhuberAI
107K Followers 0 Following Invented principles of meta-learning (1987), GANs (1990), Transformers (1991), very deep learning (1991), etc. Our AI is used many billions of times every day.Jeremy Howard @jeremyphoward
222K Followers 5K Following 🇦🇺 Co-founder: @AnswerDotAI & @FastDotAI ; Hon Professor: @UQSchoolITEE ; Digital Fellow: @StanfordEric Jang @ericjang11
69K Followers 3K Following physical AGI at 1X. Author of "AI is Good for You" https://t.co/eFg4WXhg0pRichard Socher @RichardSocher
101K Followers 970 Following CEO @youSearchEngine Investing at @aixventuresHQ Before: Stanford Adj Prof in AI/NLP, Chief Scientist at Salesforce, MetaMindKevin Patrick Murphy @sirbayes
42K Followers 334 Following Research Scientist at Google Brain / Deepmind. Interested in Bayesian Machine Learning.Kosta Derpanis @CSProfKGD
48K Followers 197 Following #CS Associate Prof @YorkUniversity, #ComputerVision Scientist Samsung #AI, @VectorInst Faculty Affiliate, TPAMI AE, #CVPR2024/#ECCV2024 Publicity Co-chairFerenc Huszár @fhuszar
40K Followers 1K Following Secular Bayesian. Associate Professor in Machine Learning @Cambridge_CL. Talent aficionado at https://t.co/RbJkoLguey Alum of @Twitter, Magic Pony and @BaldertonHorace He @cHHillee
23K Followers 449 Following Working at the intersection of ML and Systems @ PyTorch "My learning style is Horace twitter threads" - @typedfemaleDavid Pfau @pfau
22K Followers 1K Following Knowledge manifests itself in radiant dreams that shimmer like the wild sun Views are my own pfau at sigmoid dot social on 🦣 https://t.co/xqtVHHVI17 on 🦋Kyunghyun Cho @kchonyc
61K Followers 2K Following a combination of a mediocre scientist, a mediocre manager, a mediocre advisor & a mediocre PC at @nyuniversity (@CILVRatNYU) & @genentech (@PrescientDesign).Oriol Vinyals @OriolVinyalsML
166K Followers 82 Following VP of Research & Deep Learning Lead, Google DeepMind. Gemini co-lead. Past: AlphaStar, AlphaFold, AlphaCode, WaveNet, seq2seq, distillation, TF.Andrew Trask @iamtrask
74K Followers 190 Following @openminedorg, @GoogleDeepMind ethics team, @OxfordUni phd candidate, @UN pet lab, @GovAI_, creator of #GrokkingDeepLearning, NALU, and sense2vecSasha Rush @srush_nlp
52K Followers 464 Following Professor, Programmer in NYC. Cornell Tech, Hugging Face 🤗 https://t.co/cZl0wTfqGzNeurIPS Conference @NeurIPSConf
111K Followers 35 Following New Orleans, Dec 10-16, 23. https://t.co/ga8aOw615g Tweets to this account are not monitored. Please send feedback to [email protected].Percy Liang @percyliang
49K Followers 408 Following Associate Professor in computer science @Stanford @StanfordHAI @StanfordCRFM @StanfordAILab @stanfordnlp | cofounder @togethercompute | PianistAmirmojtaba Sabour @amsabour
66 Followers 40 Following Computer Science PhD student at @UofT Research Intern at @NVIDIAMarc Andreessen 🇺�.. @pmarca
1.4M Followers 24K Following Techno-optimist. E/acc. Technology brother. Move Fast and Make Things. p(Doom) = 0; p(“1984”) = not 0.Controllable Video Ge.. @cvgworkshop
41 Followers 40 Following The best time to control your video generation is today. The next best time is at ICML2024 in Vienna, Austria.Jonathan Godwin @jgodwin_ai
91 Followers 231 Following CEO at Orbital Materials. Formerly DeepMind researcher. Interested in why we aren't making more scientific progress.Rajko Radovanović @rajko_rad
4K Followers 4K Following AI/infra @a16z (partner to amazing teams eg @MistralAI @udiomusic); Enjoy most things outdoors, care about democracy in 🇷🇸🇭🇷🇸🇮🇧🇦🇲🇪justinkchen @justinkchen
986 Followers 2K Following Product @udiomusic / Ex-HEIR / Ex-Airbnb / MS @Stanford / BS @UCBerkeley Be Water, My FriendAndrew Sanchez @avincentsanchez
312 Followers 142 Following COO and Co-Founder at @udiomusic | Oxford DPhil | Harvardudio @udiomusic
28K Followers 0 FollowingEric @ericmitchellai
4K Followers 485 Following I like AI & music. Working on making LLMs easier & safer to use. Final year PhD student at Stanford advised by Chelsea Finn & Chris Manning.Alex Tomala @a__tomala
1K Followers 115 Following Research Engineer @GoogleDeepMind It’s time to ship🫡ieva @HyperboIeva
10K Followers 772 Following Quantum information, useless information, generally informed. Achieved a desperate PhD @QOQMS. Quantum algorithms researcher @PhasecraftLtdDavid Stutz @davidstutz92
3K Followers 1K Following Research scientist @DeepMind working on robust and safe AI, previously @maxplanckpress, views my own.Janne Kontkanen @jannekontkanen
161 Followers 88 Following AI & Graphics & Vision at Google ResearchHiggsfield AI @higgsfield_ai
2K Followers 9 Following We're a video AI company dedicated to democratizing social media content creation to everyone. Download Diffuse: https://t.co/sAobAz7KFTAndrew Campbell @AndrewC_ML
424 Followers 91 Following Machine Learning PhD student - Dept. Statistics University of OxfordAaron Defazio @aaron_defazio
6K Followers 363 Following Research Scientist at Meta working on optimization. Fundamental AI Research (FAIR) teamEEML @EEMLcommunity
3K Followers 11 Following Strengthening the Eastern European ML community and improving diversity in the field. https://t.co/34QAbYBeDoGeorge Morgan @vr4300
709 Followers 335 Following Fighting scale with symbols. Founder @symbolica. Previously Autopilot AI @tesla.Bruno Gavranović @bgavran3
7K Followers 954 Following Category Theory + Deep Learning Principal Scientist @symbolica bgavran.ethPreetum Nakkiran @PreetumNakkiran
10K Followers 2K Following ML research @Apple. @sh_reya’s fiancé | PhD @Harvard, postdoc @UCSanDiego, EECS @Berkeley_EECS, "AI" @OpenAI, @GoogleAIJonathan Heek @JonathanHeek
234 Followers 5 FollowingSharad Vikram @sharadvikram
1K Followers 510 Following Researcher @ Google Deepmind. I work on JAX + Pallas (https://t.co/lPMsq3yzgL) and Gemini. In the past I worked on Oryx and TFP. I like learning.Ashley Edwards @ashrewards
485 Followers 200 Following Research scientist @GoogleDeepMind. Past: Uber AI Labs, Georgia TechLeo Gao @nabla_theta
5K Followers 337 Following Alignment researcher. cofounder & head of alignment memes @ EleutherAI. currently RE @ OpenAI. Let's make the future awesome.Zhengyang Geng @ZhengyangGeng
618 Followers 588 Following PhD student @SCSatCMU with @zicokolter / Prev. Intern @Meta / Curiosity&Love / Dynamics to ASIEdward Hughes @edwardfhughes
713 Followers 403 Following #OpenEndedness. Staff Research Engineer @GoogleDeepMind, Visiting Fellow @LSEnews, Advisor @coop_ai, Choral Director @GodwineChoir. Views my own.Peter J. Liu @peterjliu
4K Followers 2K Following Research Scientist @ Google B̵r̵a̵i̵n̵ DeepMind, frontier language models research (aka chatbot engineer). Opinions are my own. 🤖🔄🚀Severi Rissanen @SeveriRissanen
43 Followers 85 FollowingOmer Bar Tal @omerbartal
2K Followers 109 Following Founding Scientist @pika_labs | ex @WeizmannScience @GoogleAITanishq Mathew Abraha.. @iScienceLuvr
54K Followers 1K Following PhD at 19 | Founder and CEO at @MedARC_AI | Research Director at @StabilityAI | @kaggle Notebooks GM | Biomed. engineer @ 14 | TEDx talk➡https://t.co/xPxwKTq6QbGuillaume Bellec @BellecGuill
937 Followers 382 Following Postdoc researcher interested in neuroscience and machine learning. Currently working at EPFL with Wulfram Gerstner.Joelle Pineau @jpineau1
10K Followers 352 Following AI researcher. VP AI Research (FAIR), @AIatMeta. Professor of Computer Science, @mcgillu. Core academic member, @Mila_QuebecCharlie Cleveland (no.. @Flayra
13K Followers 535 Following Game Director @Moonbreaker @Subnautica, @NS2. Co-founder @UnknownWorlds. Subscribe to weekly game design essays @ https://t.co/5A0qcAyDok 🏳️🌈Matthew Leavitt @leavittron
2K Followers 778 Following Chief Science Officer, Co-Founder @datologyai. Former: Head of Data Research @MosaicML; FAIR. 🧠 and 🤖 intelligence // views are from nowheremanifest ai @manifest__ai
153 Followers 0 FollowingJason Baldridge @jasonbaldridge
10K Followers 1K Following Research scientist at Google in Austin working on grounded language understanding. [email protected]Ethan @Ethan_smith_20
3K Followers 687 Following a boy and his gpu vs the world. directing research at @leonardoai_. learning as I go. uf psych. generative models and representation learningLeonardo.Ai @LeonardoAi_
43K Followers 1K Following Leonardo is a Generative AI content production suite. Create an account: https://t.co/Wn0zqQ4XDj API Access: https://t.co/LfYpTC3qiL Discord: https://t.co/Fu73yaMYNYDo models need to reason in words to benefit from chain-of-thought tokens? In our experiments, the answer is no! Models can perform on par with CoT using repeated '...' filler tokens. This raises alignment concerns: Using filler, LMs can do hidden reasoning not visible in CoT🧵
The word 'bairn' (child), which is used in Scots, Northern and Scottish English, is closely related to 'born' and 'to bear'. These words all come from a root meaning "to carry". When a baby is born it's been carried to term. The infant is then carried around. Zoom in for more:
@sedielem as mentioned i dont think it solves exposure bias. if we let previous output be the condition, we still dont have any guarantee that the actual condition will not be OOD,
any literature on diffusion being a means of avoiding error accumulation in recurrent functions? By nature it seems like its always pointing back towards the data manifold, having some self correcting properties, however i dont think this would necessarily solve exposure bias
@sedielem Re "chaotic" pieces 👇🏼. A bit derailing but illustrates nicely how smooth the overlap is :) phronesistrio.bandcamp.com/album/life-to-…
@sedielem Ah very nice stuff! Goes straight to my list. I answer with 👇🏼 (if you don't know already): on.soundcloud.com/W6b53
Nothing better than listening to technical Death Metal while focusing on lengthy partial derivatives.
In case you are wondering, this paper proves that, in general, diffusion models do not define optimal transport maps. The proof is not straightforward though (diffusion maps are optimal maps in 1D, for radial measure and for Gaussians ...) cvgmt.sns.it/media/doc/pape…
New Men out, on the mysteries of Phantom Islands. Can't believe we'd not talked about this before now, especially as Google maps fell for one. Linky link belowy-o @jayforeman = the other men. youtu.be/PVemGumEEgo?si…
@StephanMandt @karsten_kreis Funnily enough, we also tried setting the schedule such that entropy reduces linearly on it. But this doesn't work too well in practice and was usually worse than standard schedules.
Great video as always! But while "big black smudge meaning error" gets the idea across, I think the details of how the smudge came to be are really neat. When you look at the ocean in Google Maps, you're not really looking at satellite photos at all. It's totally fake. 1/🧵
New episode of #MapMen with me and @markcooperjones Why did Google Maps have a big black smudge before 2012? And why did it disappear? And what does it have to do with Captain Cook? And what is a phantom island? Share and enjoy! 🗺️ youtu.be/PVemGumEEgo?si…
New episode of #MapMen with me and @markcooperjones Why did Google Maps have a big black smudge before 2012? And why did it disappear? And what does it have to do with Captain Cook? And what is a phantom island? Share and enjoy! 🗺️ youtu.be/PVemGumEEgo?si…
@sedielem @StephanMandt @amsabour in fact tried to optimize the stepping schedules for constant rate entropy reduction (for image data), and it didn't work too well -- in line with your intuition here, @sedielem!
'Duke' comes from Latin 'dux' (leader). It's related to 'dūcere' (to lead; to pull), whence '-duce', e.g. 'to seduce' (i.e. to lead astray). The 2nd part of German 'Herzog' (duke) is cognate to 'dux'. It's related to 'ziehen' (to pull), cognate of 'dūcere'. Old English ... /1
📢📢Most diffusion (and flow matching) models use handcrafted schedules for their denoising steps during sampling. We show how to optimize them in a principled manner for high-quality generation! @amsabour added quickstart guide & collab to get you started quickly (links below)!
📢📢 Align Your Steps: Optimizing Sampling Schedules in Diffusion Models research.nvidia.com/labs/toronto-a… TL;DR: We introduce a method for obtaining improved sampling schedules for diffusion models, resulting in better samples at the same computation cost. (1/5)
Fascinating results! Is there a simple-to-understand physical criterion behind selecting the sampling schedule--keeping entropy reduction at a constant rate, @karsten_kreis ?
@amsabour Project page: research.nvidia.com/labs/toronto-a… arXiv: arxiv.org/abs/2404.14507 Quickstart guide: research.nvidia.com/labs/toronto-a… Collab: colab.research.google.com/drive/1cIwbbO4… w/ the amazing @amsabour & @FidlerSanja Big kudos in particular to the brilliant @amsabour for all the heavy lifting! 🔥
@StephanMandt @karsten_kreis Author here. Simplistically, the loss function is the difference between the ideal denoising path vs. a piece-wise linear approximation of it, weighted by the noise level (higher noise -> less weight). So a good schedule minimizes the expected "curvature" during denoising.
I think I misunderstood this. I don’t mind if you say a network infers instead of predicts. I DO mind if you say the network inferences something. “Inferencing” is not a word, it hurts my brain, even though I approve of the substring “ferenc” in there.
I will never get over how AI/ML people use the word “inference”
guy who plays piano: what's your favorite programming language guy who works at microsoft: look between C and D on your keyboard