📣 Our spicy ICML 2025 position paper: “Graph Learning Will Lose Relevance Due To Poor Benchmarks”. Graph learning is less trendy in the ML world than it was in 2020-2022. We believe the problem is in poor benchmarks that hold the field back - and suggest ways to fix it! 🧵1/10

🌟 Announcing the 2nd Workshop on Reliable and Responsible Foundation Models (R2-FM) at @icml_conf 2025 (July 13–19, Vancouver)! 📣 We welcome submissions! Submit your work here: openreview.net/group?id=ICML.… 🗓️ Deadline: May 29th, 2025 (AOE) 🔗 Website: r2-fm.github.io 💬

Bill Clinton invested $3B in 1990 for the first de novo human genome ~95%. In 2022, T2T achieved the first 100% assembly. Our AI assembler is the first of its kind - demonstrating this challenge can be solved w/ more data and compute. A call to industry: let's finish the job!

"RL with only one training example" and "Test-Time RL" are two recent papers that I found fascinating. In the "One Training example" paper the authors find one question and ask the model to solve it again and again. Every time, the model tries 8 times (the Group in GRPO), and

🎉 Congratulations to the FlashInfer team – their technical paper, "FlashInfer: Efficient and Customizable Attention Engine for LLM Inference Serving," just won best paper at #MLSys2025. 🏆 🙌 We are excited to share that we are now backing FlashInfer – a supporter and

Sigh, it's a bit of a mess. Let me just give you guys the full nuance in one stream of consciousness since I think we'll continue to get partial interpretations that confuse everyone. All the little things I post need to always be put together in one place. First, I have long

🚀 New Paper at #icml2025🚀 Diffusion models have succeeded in handling data of various modalities, such as 🖼️images (continuous), 💬languages (discrete), and 🌐manifolds (Riemannian), etc. 🤔 But how about data of 𝐦𝐢𝐱𝐞𝐝-𝐦𝐨𝐝𝐚𝐥𝐢𝐭𝐲? We investigate this question in

Hey, we built a Graph Foundation Model at Google and it's showing some very promising results! Read more in the blogpost and also catch me and Bryan Perozzi at the ICML Expo Talk next Monday. Happy to carry the Graph Learning flag ⛳️

I had a feeling, based on my past years of AI R&D experience in both industry and academia, that the AI technology stack evolves at an increasingly faster rate as the AI technology itself advances. This 'feeling' now has a formal definition - AI 4 S̶c̶i̶e̶n̶t̶i̶f̶i̶c̶AI Research

Introducing Eigent — the first multi-agent workforce on your desktop. Eigent is a team of AI agents collaborating to complete complex tasks in parallel. It is your long-term working partner with fullly customizable workers and MCPs. Public beta available to download for MacOS,

We are calling for reviewers! forms.gle/oFrpazyy2WKFGF… Spread to the expert in generative modeling and probabilistic inference!

Some advice to anyone starting a PhD in ML, or things that I heard from more experienced researchers and I tried to follow: 1) focus on a real problem. Something tangible, that can benefit people. Talk to industry folks if you're looking for open problems. Talk to the end (1/8)

🚀 Excited to announce day-0 support from NVIDIA AI Developer for OpenAI's gpt-oss model in flashinfer v0.2.10! github.com/flashinfer-ai/… ✅ Speed-of-light Blackwell mxfp4/mxfp8 MoE kernels + attention-sink from trtllm-gen ✅ FA2/FA3 template-based attention-sink support for earlier

Xin Eric Wang PC of the DB track here! Please kindly remind the reviewer that contributing new methods is not necessary for the DB track and ask the AC to take a closer look in a private message. If they don’t reply by the end of the discussion, shoot us an email and we can ping them/check too

Georgia Tech AI4Science Center is soft launched, and I'm excited to be an Associate Director. ai4science.ai.gatech.edu Collaboration+Participation of all kinds are welcomed. Please get in touch! Thanks to gtsciences for supports. Retweets appreciated! Georgia Tech #AI4Science

Thinking about model generalization is quite painful. We observe empirically that models trained with SGD on cross-entropy generalize, instead of just memorize the train data. Even when they have sufficient capacity to memorize. We do not---i repeat--- we. do. not. have a

first i thought scaling laws originated in OpenAI (2020) then i thought they came from Baidu (2017) now i am enlightened: Scaling Laws were first explored at Bell Labs (1993)

The challenge for starting agent RL research is that very few are willing to do the less glamorous but essential work. Students I worked with usually want to dive straight into training agents or experimenting with RL algorithms. They want to invent the most beautiful new “PPO,

Excited to share this new paper led by my intern Neil Kale on evaluating the adversarial robustness of the monitoring system in detecting misbehavior from autonomous agents' trajectories (e.g., CoTs + actions). It is a quite long paper with detailed setup and many empirical

Excited to teach Advanced NLP at CMU again this semester! Slides are on the course page as the course proceeds: cmu-l3.github.io/anlp-fall2025/ Lectures will be uploaded to Youtube: youtube.com/playlist?list=…