🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Roland Memisevic Mohammad Norouzi Max Welling a blue elephant riding a unicycle on the moon #imagen

It was 16 years ago, in 2006, that Geoffrey Hinton et al released their demo of deep belief nets. Undergrad me was highly impressed, and helped convince me that deep learning was the way to go. I refreshed Geoff's website almost every day checking for new papers... (1/n)

If you have any kind of feedback for the #imagen team, we'd love to hear it! imagen.research.google Chitwan Saharia, William Chan (陳俊樂), Saurabh Saxena, Lala Li, Jay Whang, Remi Denton, Kamyar Ghasemipour, Burcu Karagol Ayan, Sara Mahdavi, @iraphas13, Tim Salimans, Jonathan Ho, David Fleet.

If you don’t think DallE-2 and Imagen are an Alexnet level moment in the machine learning world you aren’t paying attention enough. Very impressive visual results coming out of these. Getting similar chills to when I saw first web browser, iPhone, etc.

Daniel has created some of the most amazing #Imagen images so far. Keep an eye out.

Happy to announce DreamFusion, our new method for Text-to-3D! dreamfusion3d.github.io We optimize a NeRF from scratch using a pretrained text-to-image diffusion model. No 3D data needed! Joint work w/ the incredible team of Ben Mildenhall Ajay Jain Jon Barron #dreamfusion

Excited to announce our work on novel view synthesis with diffusion models! Our model can lift a single 2d image into 3d. 3d-diffusion.github.io Joint work w/ William Chan (陳俊樂) Ricardo Martin-Brualla Jonathan Ho Andrea Tagliasacchi 🇨🇦 Mohammad Norouzi

Excited to announce Imagen Video, our new text-conditioned video diffusion model that generates 1280x768 24fps HD videos! #ImagenVideo imagen.research.google/video/ Work w/ William Chan (陳俊樂) Chitwan Saharia Jay Whang Ruiqi Gao Alexey Gritsenko Durk Kingma Ben Poole Mohammad Norouzi David Fleet Tim Salimans

We've just released the first version of our Deep Learning Tuning Playbook! This is our attempt to distill our process for actually getting good results with deep learning. We emphasize hyperparameter tuning since it has been a large pain point. github.com/google-researc…

Awesome work by Hshmat. The work showcases yet another evidence of how effective noise conditioning augmentation can be.

Monocular Depth Estimation using Diffusion Models abs: arxiv.org/abs/2302.14816 project page: depth-gen.github.io

.Kevin Clark & I are excited to share our new work on studying Imagen by evaluating it as a zero-shot classifier! Highlights include Imagen achieving SoTA on Stylized Imagenet and being able to perform attribute binding in certain settings unlike CLIP arxiv.org/abs/2303.15233 🧵👇

📄 An updated paper describing 3D Flexible Refinement is now out in Nature Methods! Paper: nature.com/articles/s4159… It describes further experimental #cryoEM results and the improved 3DFlex method that was released in #CryoSPARC v4.1 ❄️⚡ Tutorial: guide.cryosparc.com/processing-dat…

DRaFT backpropagates the reward directly into LoRA parameters – we don’t need to use RL because diffusion sampling is differentiable. We improve efficiency by truncating the BPTT; even truncating to one step still works! (2/5)

Our best variant, DRaFT-LV, learns 2x faster than ReFL (arxiv.org/abs/2304.05977) and 100x faster than RL. (3/5)

We can reduce the effect of the fine-tuning or mix different reward functions post-training simply by scaling down or mixing LoRAs. (4/5)

Why use RLHF when u can backprop through the sampler with any differentiable reward🥹🥹🥹

🎉🎉Excited to share that “CryoSPIN” has been accepted to #NeurIPS2024! Special thanks to my amazing collaborators and supervisors David Lindell, Marcus Brubaker, and David Fleet Paper: arxiv.org/abs/2406.10455 Webpage: shekshaa.github.io/semi-amortized… Code is also released! 🧵[1/7]

We compare our method with the state-of-the-art on both synthetic and experimental benchmarks. Empirically, cryoSPIN outperforms in reconstruction quality and FSC. [6/7]