Twitter Beğeni Hilesi

Excited to share our work on Neural Assets: a new method for enabling 3D asset-level control in image diffusion models – scalable & without any 3D inductive biases. Neural Assets goes beyond text or pixel-based control & provides an interface inspired by 3D graphics tools. 🧵

Go chat with Jackson, he is great!

2. jackson petty's Google Student Researcher project (w/ Sjoerd van Steenkiste Fei Sha Ishita Dasgupta and Dan Garrette) on how transformers LMs' depth affects their compositional generalization (it does, but they don't need to be as deep as you might think)! arxiv.org/abs/2310.19956

Jackson is presenting our work on “The Impact of Depth on Compositional Generalization in Transformer Language Models” at #NAACL today. arxiv.org/abs/2310.19956

Tiwa will be presenting our work on “A Systematic Comparison of Syllogistic Reasoning in Humans and Language Models” at #NAACL. We report variable ordering effects and more. Do check it out! arxiv.org/abs/2311.00445

possibly my favorite part of this paper is where we used Mental Models Theory, a symbolic model from cognitive psychology, to analyze LLMs' reasoning behavior

Interesting! It also suggests that LLMs can overcome this initial “tokenization bias” and learn a new grouping that is easier to work with. What prevents using char-level tokenizers then? Is it just that learning to group is exponentially harder when starting from smaller units?

We're excited to release TAPVid-3D: an evaluation benchmark of 4,000+ real world videos and 2.1 million metric 3D point trajectories, for the task of Tracking Any Point in 3D!

Glad to have been able to contribute a little to this timely and pressing research question!

Greetings from #ICML2024 in Vienna, the world's most liveable city. Check out our 5 ICML papers (2 oral), on language agents as optimizable graphs, analyzing programs (= weight matrices) of neural networks, planning & chunking & auxiliary delays for faster reinforcement learning:

Introducing Neural Jacobian Fields, robot 3D kinematic models learned only from vision! They can model & control robots from just a single RGB camera, even those w/ intractable kinematics & no embedded sensors such as soft, 3D-printed pneumatic hands! sizhe-li.github.io/publication/ne… 1/n

The application is now open for our #KempnerInstitute Research #Fellowship! Postdocs studying the foundations of #intelligence or applications of #AI are encouraged to apply. Learn more and apply by Oct. 1: bit.ly/4djPSf8 #LLMs #NeuroAI #ML Sham Kakade Bernardo Sabatini

We are now accepting nominations for reviewers and ACs at ICLR 2025. To nominate yourself or someone else, please complete this form: forms.gle/VKmG1DJgWzKTY9…

Really excited to share that I'm hiring for a Research Scientist position in our team! If you're interested in the kind of cognitively-oriented work we've been doing on learning & generalization, data properties, representations, LMs, or agents, please check it out!

Great lineup this year!

Just finished writing 26 reminders as an AC for the #NeurIPS D&B track. If you are reviewing, please make sure to respond to the author rebuttal before the discussion period ends!

I'm looking for more reviewers (especially in optimization theory) for Transactions on Machine Learning Research, please reach out with your email (you can send it to me in DM) if you're interested in becoming a reviewer! TMLR has the best system I know: 1. You don't get assigned more than 1 submission at a