🎉 BEAST3D is a #NeurIPS2026 Oral!

Can a moving animal's 3D shape predict what its neurons are doing, better than tracked keypoints? On average, yes.

BEAST3D is a self-supervised vision transformer that reconstructs animals in 3D from unlabeled multi-camera lab video. 🧵

1/7 🧵 🚨New Paper accepted at #NeurIPS2026: PA Score

Should rewording change an LLM’s uncertainty? We tackle this through conformal score design.

📄 Paper: arxiv.org/abs/2610.04239
💻 Code: github.com/Raina-Xin/PA...
📖 Blog: raina-xin.github.io/research-blo...

So I'm a #NeurIPS2026 "Top Reviewer"!
I felt I just did my job but as a physicist, (not former), this was not on my bucket list.

Our paper on Spherical Boltzmann machines is accepted at #NeurIPS2026.

A solvable energy-based model where training-time phase transitions explain sampling temperature tuning, double descent, tempered posteriors, and more.

Excited to share that our lab will be at #NeurIPS2026 with 2 papers! 🎉

Our first NeurIPS submissions since 2015, and both got in on the first try: one on temporal point processes, one on counterfactual queries on tissue graphs.

Paper threads coming soon. See you in Paris & Sydney! 👋

📢We are happy to announce that three paper by members of our department have been accepted at
@neuripsconf.bsky.social .

#NeurIPS2026

Congratulations to all authors!👏
Authors and their papers below👇

🚨SciConBench accepted at #NeurIPS2026!

Paper: arxiv.org/abs/2606.11337

Can AI *synthesize* scientific conclusions in health? We find that, on medical conclusions, we are far from it!

Check out our live dashboard: sciconbench.cs.princeton.edu

led by @hayoungjung.bsky.social

Excited to share SCION, accepted to #NeurIPS2026!

Our world is full of repetition, yet 3D reconstructions don't use it. SCION learns reusable 3D primitives + their placement, enabling photorealistic ~1 MB scenes that are easy to edit and animate.

Link: light.princeton.edu/SCION

I'm excited to share that our paper "Default Feature Representations of the Cognitive Map" has been accepted to #NeurIPS2026! 🧵

Even GPT-6 Astra gets better at spatial reasoning—without extra training.

SpatialClaw, accepted at #NeurIPS2026 🎉
11 open-source & proprietary models. All improved in average score.

💻Code: github.com/NVlabs/Spati...
📄Paper: arxiv.org/abs/2606.13673

Try it; star if useful ⭐
#papersky

In our #NeurIPS2026 paper “A Minimal Interpretable Architecture for Zero-Shot Reconstruction of Dynamical Systems” (preprint: arxiv.org/abs/2607.14937) we reduce a DS FM to ingredients minimally necessary to reproduce long-term stat. and geom. properties of DS, even outperforming many TS & DS FM.

Now this is accepted by #NeurIPS2026! As being a PI, this is the first paper for me to support a student to complete. I really love the process to discuss and polish an idea and paper with a student. This owes to Xianliang's productivity a lot.

In our #NeurIPS2026 paper “Topological Out-of-Domain Generalization in Dynamical Systems (DS) Reconstruction” (arxiv.org/abs/2606.22969) we aim to infer the DS generating observed TS jointly with control parameters, in order to extrapolate to different dynamical regimes, e.g. across tipping points.

Can LLMs reuse memory learned by another model? 🧠

Mingyuan Li presents cross-model memory transfer (#NeurIPS2026) and MemoryATHENA at the TurkuNLP Seminar.

🗓 Wed Oct 7, 13:00 EEST 📍 1171 TU3, Maarintie 8, Aalto / Zoom 🔗 www.olaresearch.org/seminar/2026...

In our #NeurIPS2026 spotlight “Parallel-in-Time Training of Recurrent Neural Networks for Dynamical Systems (DS) Reconstruction” (preprint: arxiv.org/abs/2605.12683) we speed up training of nonlinear RNNs on time series from chaotic DS by >100x by combining DEER with generalized teacher forcing.

Our #NeurIPS2026 paper borrows a 1913 finding from social psychology, the Ringelmann effect, to measure how much LLM agent teams really gain from adding agents. 🧵

#NeurIPS2026 #HAIAlignment #BiAlign
So happy to share that our paper was accepted to the BiAlign workshop at NeurIPS 2026! 🥳 Really grateful to the reviewers for such thoughtful, thorough reviews. Thank you for putting so much care into reading our work!!! 💗

1️⃣ XMemTransfer is heading to #NeurIPS2026! 🎉 We introduce "Cross-Model Memory Transfer via Target-Side Reader Adaptation" to enable seamless knowledge sharing across different models. 🔗 olaresearch.org/XMemTransfer

[3/3] Many thanks to my co-authors Weronika Kłos and Gabriel Dernbach for the great collaboration, and to @tuberlin.bsky.social, @bifold.berlin, Aignostics and Charité for their support!

Looking forward to presenting our work at NeurIPS!

#NeurIPS2026

Actually there was some more to explore 😍🎉🎉
See you at the conference! @neuripsconf.bsky.social
#Neurips2026