diack.bsky.social - Profile | ThreadSky | a Reddit-style client for Bluesky

#NewPaper We designed an "algorithmic microscope" & called it the Vendiscope. It is a new companion in the discovery process for scientists across fields and a powerful tool for diagnosing datasets and models for AI researchers. #AlgorithmicMicroscopy Link to paper: arxiv.org/abs/2502.10828

submitted 12 hours ago • 0 comments

Protein organ clocks, as assessed in over 6,000 people (~4,000 plasma proteins, 8 organs) with 20-year follow-up are predictive for over 30 age-related diseases www.thelancet.com/journals/lan...

submitted 1 day ago • 7 comments

MASt3R-SLAM code release! github.com/rmurai0610/M... Try it out on videos or with a live camera Work with @ericdexheimer.bsky.social, @ajdavison.bsky.social (Equal Contribution)

submitted 1 day ago • 2 comments

This week was the 1st publication using our immune cell sequencing to accurately make medical diagnoses. It has enormous implications, as I've reviewed here erictopol.substack.com/p/the-first-... @scottboydlab.bsky.social @science.org @anshulkundaje.bsky.social @maximzaslavsky.bsky.social

submitted 4 days ago • 8 comments

Michael Tschannen, Alexey Gritsenko, Xiao Wang, Muhammad Ferjad Naeem, Ibrahim Alabdulmohsin, Nikhil Parthasarathy, Talfan Evans, ... SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features https://arxiv.org/abs/2502.14786

submitted 5 days ago • 0 comments

Google DeepMind released SigLIP 2, open-weight vision-language encoders trained on 109 languages (incl. Swahili!). Improved multilingual capabilities, image recognition, and document understanding. Model: github.com/google-resea... Paper: arxiv.org/pdf/2502.14786 Demo: huggingface.co/blog/siglip2

submitted 5 days ago • 0 comments

SMOL is open source (CC-BY-4) and includes parralel translation of 50 African languages 🎉. Excited to see how the community uses this.

submitted 6 days ago • 0 comments

😼SMOL DATA ALERT! 😼Anouncing SMOL, a professionally-translated dataset for 115 very low-resource languages! Paper: arxiv.org/pdf/2502.12301 Huggingface: huggingface.co/datasets/goo...

submitted 7 days ago • 2 comments

AI provides a universal framework that leverages data and compute at scale to uncover higher-order patterns Today, @arcinstitute.org in collaboration with Nvidia releases Evo 2—a fully open source biological foundation model trained on genomes spanning the entire tree of life.

submitted 7 days ago • 2 comments

I just pushed a new paper to arXiv. I realized that a lot of my previous work on robust losses and nerf-y things was dancing around something simpler: a slight tweak to the classic Box-Cox power transform that makes it much more useful and stable. It's this f(x, λ) here:

submitted 8 days ago • 2 comments

PaliGemma 2 mix is out! This model can now handles short/long captioning, OCR, image Q&A, object detection, and segmentation. Available in 3B, 10B, and 28B parameter sizes and 224px/448px resolutions. Frameworks: Hugging Face Transformers, Keras, PyTorch, JAX, and Gemma.cpp. goo.gle/4i1jOOU

submitted 7 days ago • 0 comments

🎓PostDoc 👨‍🎓Habilitation position in modern geoinformatics (e.g., spatial ML, hybrid AI) for analyzing and predicting environmental changes. 📝Contract: 2+X years. 📆Apply by March 11. 🔗: bit.ly/giscience @alexanderbrenning.bsky.social, @ellis.eu, @uni-jena.de #researchteaching

submitted 11 days ago • 0 comments

Delighted to be a minor co-author on this work, led by Pranav Nair: Combining losses for different Matyroshka-nested groups of bits in each weight within a neural network leads to an accuracy improvement for models (esp. 2-bit reps). Paper: "Matryoshka Quantization" at arxiv.org/abs/2502.06786

submitted 15 days ago • 3 comments

🚨 WeatherBench Update 1. WeatherBench-X, our new evaluation code, is now on GitHub: github.com/google-resea... 2. New models (plus other small updates) on the WeatherBench website: sites.research.google/weatherbench/ 1/n

submitted 13 days ago • 1 comment

AI in action just as I am leaving Paris.

submitted 13 days ago • 2 comments

Announcing the release of Common Corpus 2. The largest fully open corpus for pretraining comes back better than ever: 2 trillion tokens with document-level licensing, provenance and language information. huggingface.co/datasets/Ple...

submitted 15 days ago • 2 comments

I spoke to France 24 about our research project and open source dataset. youtu.be/9VtoKJWZ5os?...

submitted 15 days ago • 0 comments

I'm in Paris and excited to visit the AI African Village this afternoon! Happy to chat about AI research and startups based in Africa. Registration is still open btw if you'd like to join: aivillage.africa #ParisAIActionSummit

submitted 15 days ago • 1 comment

Maybe a hot take, but what about the following advice to the next gen: Don't get an AI degree; the curriculum will be outdated before you graduate. Instead, study math, stats, or physics as your foundation, and stay current with AI through code-focused books, blogs, and papers.

submitted 17 days ago • 12 comments

Paris reminds me of NYC a couple of years ago: electric bikes everywhere, I even saw electric rollerblades! The fascinating part for me is how efficient these small batteries are getting

submitted 17 days ago • 0 comments

We've built a simulated driving agent that we trained on 1.6 billion km of driving with no human data. It is SOTA on every planning benchmark we tried. In self-play, it goes 20 years between collisions.

submitted 20 days ago • 22 comments

Making LLMs run efficiently can feel scary, but scaling isn’t magic, it’s math! We wanted to demystify the “systems view” of LLMs and wrote a little textbook called “How To Scale Your Model” which we’re releasing today. 1/n

submitted 22 days ago • 3 comments

I am deeply honored to receive the #QEPrize2025 from @qeprize.bsky.social alongside such an exceptional group of colleagues, whose groundbreaking scientific contributions have had a profound impact on the field of Machine Learning. qeprize.org/winners/mode...

submitted 22 days ago • 2 comments

Congrats to @adjiboussodieng.bsky.social on being named an Early-Career Distinguished Presenter at the MRS Spring Meeting 2025! This honor is meant to highlight exciting new research directions in materials science. 🐯 #ChemScky #Princeton #VendiScoring #MatSci www.cs.princeton.edu/news/adji-bo...

submitted 22 days ago • 0 comments

🚀 Registration Now Open for the @ELLISforEurope Summer School: AI for Earth & Climate Sciences! 🌍🤖 Join us in Jena, Germany from September 1–5, 2025 📅 Apply now through March 31. Notifications by April. 🔗 www.ellis-jena.eu/summer-schoo... 🎯 @uni-jena.de, @MPI-BGC, @carlzeissstiftung.bsky.social

submitted 23 days ago • 0 comments

If you are interested in developing large-scale, multimodal datasets & benchmarks, and advancing AI through data-centric research, check out this great opportunity. Our team is hiring! boards.greenhouse.io/deepmind/job...

submitted 32 days ago • 0 comments

Arsenal is looking for a research engineer because they need machines to win trophies 😁 careers.arsenal.com/jobs/5434108...

submitted 33 days ago • 1 comment

um

submitted 37 days ago • 7219 comments

TRecViT: A Recurrent Video Transformer arxiv.org/abs/2412.14294 Causal, 3× fewer parameters, 12× less memory, 5× higher FLOPs than (non-causal) ViViT, matching / outperforming on Kinetics & SSv2 action recognition. Code and checkpoints out soon.

submitted 47 days ago • 1 comment

🎉 50,000+ annotations reached! The FineWeb2-C community is helping build better language models on annotation at a time. 📊 Current stats: - 115 languages represented - 419 amazing contributors - 24 languages with complete datasets But we're not done yet! 🧵

submitted 41 days ago • 1 comment

We’re excited to introduce Transformer², a machine learning system that dynamically adjusts its weights for various tasks! sakana.ai/transformer-... Adaptation is a remarkable natural phenomenon, like how the octopus blends into its environment, or how the brain rewires itself after injury. 🧵 1/N

submitted 42 days ago • 1 comment

If you use generative #DeepLearning for molecule design, check out our latest work, where we perform a large scale analysis (~1 B designs!) and find ‘traps’, ‘treasures’ and ‘ways out’ in the jungle of generative drug discovery. 🌴 🐒 Paper: arxiv.org/abs/2501.05457 Code: github.com/molML/jungle...

submitted 43 days ago • 0 comments

Helium 2B running locally on an iPhone 16 Pro at ~28 tok/s, faster than you can read your loga lessons in French 🚀 All that thanks to mlx-swift with q4 quantization!

submitted 43 days ago • 0 comments

An interesting week of small models.

submitted 43 days ago • 1 comment

Paris Open Source AI Summit https://opensource.org/events/paris-open-source-ai-summit

submitted 43 days ago • 0 comments

🚨📢 Call for short and tiny Papers 🚨📢 for the third Machine Learning for Remote Sensing workshop at ICLR 2025 in Singapore! Deadline in 3 weeks on February 3rd! Check the website ⬇️ for submission details ml-for-rs.github.io/iclr2025/ @iclr-conf.bsky.social

submitted 44 days ago • 1 comment

Mastodon announces transition to non-profit structure

submitted 44 days ago • 2 comments

Took down my last post. Assistive technology and AI models... it's something we need to think carefully about... Still, gaze detection in such a small and open model as Moondream is fascinating! Moondream 2.0: huggingface.co/vikhyatk/moo... Blog: moondream.ai/blog/announc...

submitted 44 days ago • 0 comments

new moondream, new me huggingface.co/vikhyatk/moo...

submitted 47 days ago • 1 comment

💙 𝗗𝗮𝘁𝗮 𝗔𝗻𝗻𝗼𝘁𝗮𝘁𝗶𝗼𝗻 𝗕𝗼𝘁𝘁𝗹𝗲𝗻𝗲𝗰𝗸 𝗮𝗻𝗱 𝗔𝗰𝘁𝗶𝘃𝗲 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝗳𝗼𝗿 𝗡𝗟𝗣 𝗶𝗻 𝘁𝗵𝗲 𝗘𝗿𝗮 𝗼𝗳 𝗟𝗟𝗠𝘀 💡 Have you ever had to overcome a lack of labeled data to deal with an NLP task? We are conducting a survey to explore the strategies used to overcome this bottleneck. #NLP #ML

submitted 73 days ago • 2 comments

Inspired by DeepMind's Astra, Project Pastra is Heiko Hotz's fascinating dev guide and code for Gemini Live Multimodal API - near realtime audio/video processing. github.com/heiko-hotz/g... Check out my walkthrough video: youtu.be/ZUY4O4_Xoe0

submitted 46 days ago • 0 comments

Google just released TimesFM-2.0 (Time Series Foundation Model - jax & pytorch) on Hugging Face with a significant boost in accuracy and maximum context length. It is a pretrained time-series foundation model developed by Google Research for time-series forecasting. huggingface.co/google/times...

submitted 47 days ago • 0 comments

Los Angeles is experiencing intense wildfires! See the affected areas on Google Maps, which includes a smoke map: www.google.com/maps/@33.980... You can read more about the research behind this here: sites.research.google/gr/wildfires/

submitted 49 days ago • 0 comments

😂

submitted 50 days ago • 0 comments

Introducing METAGENE-1🧬, an open-source 7B-parameter metagenomics foundation model pretrained on 1.5 trillion base pairs. Built for pandemic monitoring, pathogen detection, and biosurveillance, with SOTA results across many genomics tasks. 🧵1/

submitted 51 days ago • 2 comments