Amos (@AmosGropp) header
Amos (@AmosGropp) profile picture

🦆

Joined January 2019
Photo by Amos (@AmosGropp)Photo by Amos (@AmosGropp)Photo by Amos (@AmosGropp)Photo by Amos (@AmosGropp)
I think this paper [arxiv.org/abs/2510.08570] wins the "strangest" (in a good sense) 1-step diffusion award of this year. They parametrize a model as an invertible network, which maps from the sample space to the representation space, which is assumed to be linear: i.e. we assume that estimating the (unconditional) velocity from a noised sample is a linear operation (which we learn as well) in that "latent" space. This can be seen as a latent diffusion model, where your encoder is a (large) invertible neural network, your decoder is its inverse, and your diffusion model is a linear function in the latent space. Quite a curious construction to be honest. Why do we need invertibility instead of training a separate "decoder" to map the "latents" back (starts smelling with vanilla LDMs here)? I find it to be just a convenient design constraint to make the theory clean (otherwise, you need to bother with upper/lower bounds and dozens of weaker assumptions everywhere). (Note: the paper presents are general idea of mapping data to a linear space, but I was mainly reading through this nice diffusion example.)
9
47
2
383
33,736
Replying to @nadavcohen
@nadavcohen and I uploaded lecture notes on the theory (and surprising practical applications) of linear neural networks. Hope that it can be useful, especially to those entering the field as it highlights distinctions between DL and "classical" ML theory arxiv.org/abs/2408.13767
3
1
16
1,461
Presenting M-SDF this week at #CVPR2024 ! Come to our poster on Wednesday morning to say hi and learn/hear about training Flow Matching generative model with our new 3D representation
Thrilled to share Mosaic-SDF (M-SDF), a simple 3D representation suitable for 3D generative models! Check out more results of text-to-3D generations here - lioryariv.github.io/msdf
11
2
84
12,078
Stable diffusion 3 uses Flow Matching! An amazing milestone ! Read more about flow matching here: arxiv.org/abs/2210.02747 @lipmanya @RickyTQChen @mnick @lematt1991
Stability AI announces Stable Diffusion 3 most capable text-to-image model, utilizing a diffusion transformer architecture for greatly improved performance in multi-subject prompts, image quality, and spelling abilities. Prompt: Epic anime artwork of a wizard atop a mountain at night casting a cosmic spell into the dark sky that says "Stable Diffusion 3" made out of colorful energy
21
126
19,664
Thanks for sharing our work! joint work with @OmriPuny, @itai_gat, Brian Karrer, @urielsinger and @lipmanya
D-Flow Differentiating through Flows for Controlled Generation Taming the generation outcome of state of the art Diffusion and Flow-Matching (FM) models without having to re-train a task-specific model unlocks a powerful tool for solving inverse problems, conditional generation, and controlled generation in general. In this work we introduce D-Flow, a simple framework for controlling the generation process by differentiating through the flow, optimizing for the source (noise) point. We motivate this framework by our key observation stating that for Diffusion/FM models trained with Gaussian probability paths, differentiating through the generation process projects gradient on the data manifold, implicitly injecting the prior into the optimization process. We validate our framework on linear and non-linear controlled generation problems including: image and audio inverse problems and conditional molecule generation reaching state of the art performance across all.
7
32
12,691
✨ Happy to share that my new blog post about Gaussian Splatting, based on a group meeting in the lab of @talidekel, was selected as an Editors' Pick by Towards Data Science ✨ w @GeyerMichal
A comprehensive overview of Gaussian Splatting - Everything you need to know about the new trend in the field of 3D representations by Kate Yurkova buff.ly/3GTNAo7
2
12
41
13,543
Thrilled to share Mosaic-SDF (M-SDF), a simple 3D representation suitable for 3D generative models! Check out more results of text-to-3D generations here - lioryariv.github.io/msdf
8
72
4
468
75,298
You didn’t believe Jewish babies were decapitated, even when there was eye witness testimony & photos. You did believe Israel was responsible for the hospital attack, with zero evidence & when it was proven it was not Israel you stayed silent. You just hate Jews.
153
1,193
56
4,234
151,283
**Viewer discretion is advised** For all those of you who deny the massacre occurring on October 7th. Don't look away. Hamas is ISIS. It's either you're against terror or you support Hamas an ISIS-like organization - an organization that causes suffering both in Israel and Gaza alike. #HamasisISIS #Hamas_is_ISIS #Israel #Gaza #Hamas
1,123
3,079
649
5,647
506,673
By now, it's crystal clear. Share #HamasisISIS
263
2,139
122
4,840
174,245
גיא פלג דייק. יש לי עשרות שעות שיחה מתועדת עם יצחק אילן, "הגרוזיני". הוא קרא לו מחבל יהודי
142
448
43
2,927
92,681
פספסתי את הנאום של נתניהו. מישהו יכול להשתין עליי?
51
77
6
1,955
38,171
בית המשפט אישר את מועמדות אפי הזנאי לתפקיד יו"ר לשכת עורכי הדין. אפי אינו רק מועמד לתפקיד ראש הלשכה, אלא המסלול עוקף רפורמה בועדה למינוי שופטים. השגרה שלי די מפנקת, אך בהינתן תמיכה ציבורית במאבק נגדו - אחזור מהפנסיה, לפחות עד סוף הבחירות. לתמיכה במאבק: drove.com/campaign/645d2bb19…
218
161
29
1,788
111,809
BREAKING: The future of object detection is here! Meet YOLO-NAS, open-sourced YOLO-based architecture redefining state-of-the-art in object detection.
60
587
97
3,044
726,704
YOLO-NAS is here! 🔥 A new open-sourced YOLO-based architecture, just dropped! And it's setting a new standards for SOTA Object Detection! 🚀 -- starter notebook in the thread -- Read more 🧵👇
13
36
6
141
37,359
בנט כעת בכאן 11: תחת הממשלה הזאת אני לא ישן בשקט. זאת הממשלה הגרועה מקום המדינה. זה היה נאום יאיר נתניהו. היה פה מופע עצוב של אדם שאיבד את זה
216
218
23
3,844
106,917
רק משפט אחד של אמת היה בנאום השקרים של נתניהו הערב, המשפט "אתם מכירים אותי". אכן מכירים אותך ולא מאמינים למילה שיוצאת לך מהפה, יא אפס. לך הביתה – at Israel
49
129
9
2,820
42,629
באנגלית זה נשמע אפילו יותר טוב. כלומר יותר רע.
100
469
46
2,311
139,850
אם יש לקח אחד שכולנו חייבים לעצמנו, זה הסיפור של פולין 🇵🇱 מלחמת ההתשה של הממשלה שם הצליחה. הציבור התעייף. מערכת המשפט הפולנית מזוהמת פוליטית היום ללא הכר. לוין ורואים נסעו לשם ללמוד את השיטה. בינם לבין הרס הדמוקרטיה נמצאת רק הנחישות שלנו
24
208
20
863
41,218
לשרשור הבא אני קורא "אלימות משני הצדדים"
132
262
50
2,105
186,259