Twitter Beğeni Hilesi

Going to present this in Vienna in May :) See you there? #ICLR2024

By popular demand, we updated the GIVT arxiv with a derivation of the loss function and additional details arxiv.org/abs/2312.02116 We're also working on the code release - stay tuned!

Very interesting work by some colleagues in Google DeepMind: Oliver and Hartmut present a new JAX library for easy and fast machine learning under E(3)-equivariance, e.g. point clouds in 3D space. paper: arxiv.org/abs/2401.07595 code: github.com/google-researc…

How is next-token prediction capable of such intelligent behavior? I’m very excited to share our work, where we study the fractal structure of language. TLDR: thinking of next-token prediction in language as “word statistics” is a big oversimplification! arxiv.org/abs/2402.01825

Super excited about this work where we improve the MusicLM text-to-music model with RL and user feedback and see substantial gains in performance.

Training a simple classifier on top of "frozen features" from large vision models is now common and powerful. In our CVPR 2024 paper (arxiv.org/abs/2403.10519), we show that just applying simple augmentations on such frozen features can improve few-shot classification.

The GIVT that keeps giving. We scaled up our soft token generation transformer and improved the GMM. Also there is 👨🏽‍💻 CODE 🧑‍💻. Check it out ✨

GiVT improvements and checkpoints! As a reminder, GiVT performs autoregressive image generation with continuous latents, (surprisingly) outperformimg the standard approach of generating discrete tokens.

Slowly but surely freeing us from the discretization in tokenization! Now the code and checkpoints of GIVT are available, so you can play around with them:

Happening today at #ICLR2024, Poster #39 at 4.30 PM! Come by :)

We release PaliGemma. I'll keep it short, still on vacation: - sota open base VLM designed to transfer quickly, easily, and strongly to a wide range of tasks - Also does detection and segmentation - We provide lots of examples - Meaty tech report later! ai.google.dev/gemma/docs/pal…

We introduced PaliGemma at Google I/O. I would like to quote a few links in this thread for developers, most importantly, share our new academic program to support PaliGemma research with Google Cloud credits🧵

Great work by Boris Dayma 🖍️ and Pedro Cuenca creating an open-source reproduction of CapPa! They present some interesting architecture tweaks including the use of registers.

✨PaliGemma report will hit arxiv tonight. We tried hard to make it interesting, and not "here model. sota results. kthxbye." So here's some of the many interesting ablations we did, check the paper tomorrow for more! 🧶