Berkeley AI Research
@berkeley_ai
We're graduate students, postdocs, faculty and scientists at the cutting edge of artificial intelligence research.
Some problems can’t be rushed—they can only be done step by step, no matter how many people or processors you throw at them. We’ve scaled AI by making everything bigger and more parallel: Our models are parallel. Our scaling is parallel. Our GPUs are parallel. But what if the…
Our code has been released: github.com/jessicayin/tac…. Check out how we simulate the magnetic tactile skin and transfer it to the real world. As I near the end of my PhD, 90% of my projects (18 out of 20) now have public code. Just two more to go! I'm grateful to all my amazing…
Introducing tactile skin sim-to-real for dexterous in-hand translation! We propose a simulation model for ReSkin, a magnetic tactile sensing skin. It can simulate ternary shear and binary normal forces. More: jessicayin.github.io/tactile-skin-r…
SANDLab had a great showing last week at the IEEE Engineering Medicine and Biology Conference (EMBC 2025) in Copenhagen! We presented 4 posters representing 3 peer-reviewed papers and 1 research abstract. @IEEEembs @UCJointCPH @berkeley_ai @BerkeleyCDSS
This Thursday 12-1pm: It's not too late to sign up.
Looking fwd to presenting this talk @Google next Thurs at noon. It will be live in person in Mountain View CA (not online) but is free and open to the public: How to Close the 100,000 Year “Data Gap” in Robotics rsvp.withgoogle.com/events/how-to-…
Can LLMs write code and learn like novice programmers? We release ParaStudent, a framework to study how to make LLMs generate realistic, student-like code, which is often imperfect, iterative, and stylistically diverse 👩🎓 Paper and code shared in the thread 👇
I wrote a fun little article about all the ways to dodge the need for real-world robot data. I think it has a cute title. sergeylevine.substack.com/p/sporks-of-agi
Join us on July 19th at @icmlconf, Vancouver, for the EXAIT Workshop— a full-day workshop on the role of exploration in AI today.
📢 Big update: Introducing BFCL V4 Agentic — and BFCL published at ICML 2025! 🏟️ Some BFCL lore... back in 2022, as researchers we couldn't find good open-source models that could handle zero-shot function calling — so we decided to train our own. Sounds simple, right? It was!…
For systems that make predictions about individuals, there's growing interest in ensuring performance for all relevant subpopulations. But who are these subpopulations? And how can we provide guarantees without knowing them in advance? Come see our #ICML2025 poster at 11AM today…
For everyone interested in precise 📷camera control 📷 in transformers [e.g., video / world model etc] Stop settling for Plücker raymaps -- use camera-aware relative PE in your attention layers, like RoPE (for LLMs) but for cameras! Paper & code: liruilong.cn/prope/
How can we train a foundation model to internalize what it means to “explore”? Come check out our work on “behavioral exploration” at ICML25 to find out!
🧠 How “old” is your model? Put it to the test with the KiVA Challenge: an abstract visual reasoning benchmark grounded in real developmental data from children (3–12) & adults. Join here: eval.ai/web/challenges… @AlisonGopnik @shiryginosar @kate_saenko_ @anisanmajhi @mqraitem
Flow Q-learning (FQL) is a simple method to train/fine-tune an expressive flow policy with RL. Come visit our poster at 4:30p-7p this Wed (evening session, 2nd day)!
Excited to introduce flow Q-learning (FQL)! Flow Q-learning is a *simple* and scalable data-driven RL method that trains an expressive policy with flow matching. Paper: arxiv.org/abs/2502.02538 Project page: seohong.me/projects/fql/ Thread ↓
We’re organizing the RoboArena Challenge at CoRL this year! Show the performance of your best generalist policy, in a fair, open benchmark for the robotics community! 🤖 Sign up, even if you don’t have a robot! More details in 🧵👇
If you are at #ICML2025, don't miss our talks/posters! - Medical LLM Benchmarking (Position Paper): Oral - July 16, 3:30 PM + Poster - July 16, 4:30 PM by @tom_hartvigsen - Venn-Abers Calibration + Conformal Prediction: Poster on July 16, 4:30 PM by @LarsvanderLaan3
.@rajivmovva and @kennylpeng will be at ICML presenting our method, HypotheSAEs, for hypothesis generation from text - go chat with them!
1. We will present HypotheSAEs at #ICML2025, Wednesday 11am (West Hall B2-B3 #W-421). 2. Let me know if you'd like to chat about: - AI for hypothesis generation - why SAEs are still useful - whether PhD students should stay in school
1. We will present HypotheSAEs at #ICML2025, Wednesday 11am (West Hall B2-B3 #W-421). 2. Let me know if you'd like to chat about: - AI for hypothesis generation - why SAEs are still useful - whether PhD students should stay in school
At #ICML25 @icmlconf this week! Come say hi 👋—happy to chat about your research, ideas, or AGI :) I’m presenting work on: - Efficient video generation (Sparse VideoGen) - Fast fine-tuning w/ sparsity (SparseLoRA) - Parallel reasoning (APR) Schedule in threads 🧵
🎉 Come check out our poster at #ICML2025! 🚀 Sparse VideoGen: Accelerating Video Diffusion Transformers with Spatial-Temporal Sparsity 📍 East Exhibition Hall A-B — #E-3307 🗓️ Poster Session 2 | Tue, Jul 15 | 🕓 4:30–7:00 PM ⚡ We speed up video diffusion transformers by over…
I'm at ICML '25! Come check out our benchmark LMRL-Gym for multi-turn RL for LLMs at Wednesday's Poster Session. In addition to dialogue & text game tasks, we share a methodology for synthetic data generation to develop RL algorithms. Paper & code here: lmrl-gym.github.io