🚀 Introducing Flextron - a Many-in-One LLM - Oral at ICML! Train one model and get many optimal models for each GPU at inference without any additional retraining. 🌟 🔗 Paper: arxiv.org/abs/2406.10260 Main benefits with only 5% post-training finetuning: ✅ Best model for

Inspired by excellent works such as Matformer and OFA, we adapted the concept for LLMs and showed its efficacy for attention and MLP. 🙌 Big thanks to all contributors: Ruisi Cai, Saurav Muralidharan, GREG, Hongxu (Danny) Yin, VITA Group, Jan Kautz #ICML2024 #LLM #MachineLearning #AI

What a wonderfully organized symposium, an apt celebration of the career of Prof. Narahari, a true role model for academics and researchers!

Excellent crop of ICML best papers. Congratulations! 🎉🎉🎉 All are great, but a couple stand out to me personally: Florian Tramèr & my Google DeepMind colleague Nicholas Carlini have TWO best papers. One with Gautam Kamath and one with me (and others). 1/

Story in "India Today" on our work with ARMMAN on using restless bandits for maternal health "helping at-risk expectant mothers stay engaged with prenatal care, reducing dropout rates" #AIforSocialImpact indiatoday.in/technology/fea…

Google presents Mixture of Nested Experts Adaptive Processing of Visual Tokens The visual medium (images and videos) naturally contains a large amount of information redundancy, thereby providing a great opportunity for leveraging efficiency in processing. While Vision

After 5 beautiful years PhDing Allen School, I joined Google DeepMind as a Research Scientist & relocated to the SF Bay Area (let us grab a coffee!) 🌉 I'm excited to build (& collaborate on) next gen ML fundamentals & models -- adaptive compute, world models, new capabilities ...!✨

We have an experimental updated version of Gemini 1.5 Pro that is #1 on the lmsys.org Chatbot Arena. This model is a significant improvement over earlier versions of Gemini 1.5 Pro (it cracks into 1300+ elo score territory). I'm really proud of the whole team of people that

We're humbled by the early positive reception and shares for our MoNE paper! 🙏 A huge shoutout to my amazing collaborators and mentors for their incredible contributions - Sujoy Paul Nidhi Hegde Aditya Kusupati Prateek Jain Arsha Nagrani Shyamal Buch, Anurag Arnab ✨

Really excited to see us talk publicly about the progress on this work. A few people in our awesome robotics team many years ago came to me and suggested that we should work on this problem (even though it would require buying some pricey higher speed robots) because it would

Exciting opportunity in the Languages team at Google DeepMind India to advance LLM frontiers and bring their benefits to a lot more people! Get in touch for any queries. I shall also be at ACL 2025 next week, happy to chat in person. boards.greenhouse.io/deepmind/jobs/…

I personally love listening to podcasts, and I'm excited that we're launching season 3 of the Google DeepMind podcast, covering some of the most important topics in AI.

Jeff Dean (@🏡) Thang Luong Quoc Le Diyi Yang Yi Tay The panel on stage..

Headed back with incredible memories from Vietnam and Singapore! A crazy fun and productive week at the GenAI Summit and FUV in Vietnam ➡️ NUS + Google office in SIN! Hanging out with Prateek Jain Jeff Dean (@🏡) Thang Luong Quoc Le Yi Tay and co! Fun + work pics below!

*Mixture of Nested Experts: Adaptive Processing of Visual Tokens* by Gagan Jain Sujoy Paul Aditya Kusupati Prateek Jain Arsha Nagrani MoE variant where experts are slices of a single MLP, allowing for dynamic compute for tokens given a budget. arxiv.org/abs/2407.19985

Matryoshka Representation Learning (MRL) is a super exciting approach to improving the quality and efficiency of embedding models and strategies ✨ MRL allows models to store more information in the earlier dimensions of their data vectors. This method not only boosts

[Life Update] I've moved to the Bay Area to lead academic research investments for Gemini! Academic friends - what can we do better for your research? DMs open! I'm keen to meet new people - let's chat about AI, HCI, foundational / academic research, startups or let's just hike

Past 7 yrs in Bangalore have been very rewarding - lots of growing up, building a new research program and helping build a new lab from the ground-up with the best people! Prateek Jain Manish Gupta Partha Talukdar Milind Tambe Shachi Dave and many more !

Prateek Jain Manish Gupta Partha Talukdar Milind Tambe Shachi Dave also very grateful to my academic friends in India, South and Southeast Asia, I'll continue to be a cheerleader for you all always :D Big thanks to Ashwani Sharma who has opened all of these doors over the years!

Check out NotebookLM! Create a notebook, upload one or more sources (e.g. PDFs of research papers, your favorite PhD thesis, a newspaper article, etc) then click on 'Generate' to create a podcast of two voices talking about the content you've uploaded. blog.google/technology/ai/…