Check out our #ICLR2024 paper, “SineNet: Learning Temporal Dynamics in Time-Dependent Partial Differential Equations” (openreview.net/forum?id=LSYhE…). U-Nets are widely used for static image segmentation tasks in computer vision. [1/N]

This is a super interesting paper, proving once more how important residual connections are. In this case residuals are interpreted as small transformations over time, strongly improving the UNet idea for spatio-temporal problems! I learned a lot when reading this :)

Looking forward to presenting our work on temporal dynamics modeling at #ICLR2024! We will be poster #30 on Friday morning from 10:45 to 12:45PM Vienna time.

Come chat with Jacob and me about SineNet and neural PDE solvers in Friday morning’s poster session!

A successful poster session of our work: Active Test-Time Adaptation at #ICLR2024! Thanks Jacob Helwig for presenting on behalf of us!🥰

Jacob Jacob Helwig and I will present our paper "SineNet: Learning Temporal Dynamics in Time-Dependent Partial Differential Equations" tomorrow (May 13) from 09:00 to 10:15 EDT in the AI4ScienceTalk series AI4Science Talks. Come and join us!. Details: ai4sciencetalks.github.io/projects/sinen…

Introducing [#ICML2024] *Minimal Frame Averaging for More Symmetries and Efficiency*, a model-agnostic framework for efficient equivariance across a wide range of groups. Joint work w/ Jacob Helwig Shurui Gui Shuiwang Ji P: arxiv.org/abs/2406.07598 C: github.com/divelab/MFA

Check out our #ICML2024 Spotlight paper on efficiently instilling equivariance in non-equivariant architectures via frame averaging. We additionally extend FA to groups not yet considered by previous FA methods, such as the Lorentz group for describing symmetries in spacetime

I am leading the new RAISE Initiative (raise-tamu.net). The aim of RAISE is to take a first step in building a coordination network within TAMU to promote collaborations among AI, science, and engineering. Very grateful to my colleagues and friends for support.

(1/n) Training LLMs can be hindered by out-of-memory, scaling batch size, and seq length. Add one line to boost multi-GPU training throughput by 20% and reduce memory usage by 60%. Introducing Liger-Kernel: Efficient Triton Kernels for LLM Training. github.com/linkedin/Liger…

Excited to share our NeurIPS paper Mat2Seq will be presented at East Exhibit Hall A-C #2402 next week Wed from 1 p.m. CST — 4 p.m. CST by Jacob Helwig ! Sad that I cannot attend NeurIPS this year, but my great supervisor Dr. Ji Shuiwang Ji will be there. #NeurIPS2024