Shubhendu Trivedi

“[Mathematics research] funding is being revoked or cut—which, according to experts, could be catastrophic for the present and future of the field.” Read more in Emma R. Hasson’s article, “Math Is Quietly in Crisis over NSF Funding Cuts,” in @sciam. Link in comments.

Some "yellow book" (i.e. "probabilistic theory of pattern recognition") treatment of conformal prediction arxiv.org/abs/2507.15741

With LLMs, one might soon need to write an elegy for the death of the crank as a human figure. Cranks were interesting in that they did not really invent nonsense, or anything at all, but possessed a certain obsession to rehearse legitimacy. Now all it takes is a set of prompts.

Are NeurIPS position track reviews released yet? Is the timeline for them different?

Want to work at the frontier of active learning and AI-guided experimentation? @EliWeinstein6 is hiring PhD students at DTU in Copenhagen.

Adaptive Transition State Refinement with Learned Equilibrium Flows 1. This novel study introduces Adaptive Equilibrium Flow Matching (AEFM), a novel AI approach that significantly enhances the accuracy of transition state (TS) predictions in chemical reactions, crucial for…

Another nice benchmarking paper highlighting the rapid exciting progress of universal MLIPS/NNPs:

Why do video models handle motion so poorly? It might be lack of motion equivariance. Very excited to introduce: Flow Equivariant RNNs (FERNNs), the first sequence models to respect symmetries over time. Paper: arxiv.org/abs/2507.14793 Blog: kempnerinstitute.harvard.edu/research/deepe… 1/🧵

Wrapped up Stanford CS336 (Language Models from Scratch), taught with an amazing team @tatsu_hashimoto @marcelroed @neilbband @rckpudi. Researchers are becoming detached from the technical details of how LMs work. In CS336, we try to fix that by having students build everything:

To understand what is happening in the RNN, we extract automata from its hidden representations during training, which visualize the computational algorithm as it is being developed. (5/11)