Puheng Li

Check this out!

The #ShawPrize in #MathematicalSciences 2026 is awarded in equal shares to Emmanuel Candès and Camillo De Lellis @Stanford @the_IAS #Shawlaureates2026

🚀 Today, we’re excited to introduce SimpleTES for scaling the scientific discovery loop. 🧵 I always ask myself: what are we actually scaling in scientific discovery? Most LLM discovery methods focus on test-time scaling generation — more tokens, more agents, more turns. But science advances through the evaluation-driven loops: propose → evaluate → refine → repeat. SimleTES captures this idea, discovering SOTA solutions across 21 scientific problems! Key discoveries: 🏎️ 2.17x faster lasso solver than glmnet — the gold-standard LASSO solver, engineered for decades. ⚛️ 24.5% fewer quantum routing overhead on IBM Q20 — superior than previous standard library LightSABRE. 📐 0.380868 on Erdős Minimum Overlap — outperforming previous solutions from mixed-frontier ensembles or humans. 🧬 0.74 on Tabula Muris (scRNA-seq denoising) — new SOTA, generalizing to unseen tissue types without retraining. #LLM #AI4Science #ScalingLaws #SimpleTES #MachineLearning

Finally getting to share one of my favorite projects. ICLR Oral! 🏆 It’s so strange how rigid video tokenization is. Think about it: why should a still landscape cost the same amount of tokens as a busy street? We built InfoTok. We went back to basics with Shannon’s information theory to make tokens "adaptive" in a principled way. Its 2.3x better compression and 11x faster inference demonstrates the magic of the old-school theory ✨ Check it out: research.nvidia.com/labs/dir/infot…

🤔Want a principled way to RL your diffusion model? Check Data-regularized Reinforcement Learning (DDRL)! Post-train @nvidia #Cosmos World Foundation models with a million GPU hours! 🤯 Novel formulation ➡️ Theoretically integrates SFT into RL ➡️ Robust to Reward Hacking 🛑 Details: research.nvidia.com/labs/dir/ddrl/ #DDRL #Diffusion #RL #NVIDIA #Cosmos

Analyzing the Role of Permutation Invariance in Linear Mode Connectivity ift.tt/ckwKPeG

I love you ChatGPT.

I had the privilege of being interviewed by the marvelous and unique Steve Strogatz on the podcast "The Joy of Why." We talked about how AI is changing the science of prediction. Links are below: open.spotify.com/episode/50Y4M6… quantamagazine.org/tag/the-joy-of… podcasts.apple.com/us/podcast/the…

Exploring Neural Network Landscapes: Star-Shaped and Geodesic Connectivity ift.tt/JFYU39P

Happy new year and all those things! :) We start it with todays reading group session about "On the Generalization Properties of Diffusion Models" arxiv.org/abs/2311.01797 with @lphLeo623 and @lhktkz! Join us at 11am ET / 5pm CET / 4pm UTC on Zoom: portal.valencelabs.com/logg

Huge congratulations!! @james_y_zou

@StanfordDBDS Huge congratulations!! @james_y_zou

We are thrilled to announce that James Zou has been promoted to DBDS Associate Professor with tenure. A hearty congratulations and thanks to his endless hard work, revolutionary research and constant contributions to our department.

🎉Fantastic collaboration with Zhong Li @lhktkz , Huishuai Zhang @HuishuaiZhang , and Jiang Bian @jbian22 at Microsoft Research! 📄openreview.net/pdf?id=hCUG1MC… #generativeai #diffusionmodels (4/4)

2. 📊Data-dependent regime: For target distributions with increasing modes distances, the generalization performance of diffusion models becomes significantly worse. (3/4)

📢Excited to share our new #NeurIPS2023 work on the generalization properties of 𝗱𝗶𝗳𝗳𝘂𝘀𝗶𝗼𝗻 𝗺𝗼𝗱𝗲𝗹𝘀! We propose a data-independent estimate of the generalization error for score-based diffusion models, and extend it to a 📊data-dependent "modes shift" scenario: (1/4)

On the Generalization Properties of Diffusion Models. (arXiv:2311.01797v1 [cs.LG]) ift.tt/62vfo4E

So happy to have presented our work on algorithmic fairness (with @james_y_zou and @linjunz_stat) at @iclr_conf !

I will be participating in #ICLR2023 in Kigali, Rwanda. Excited to introduce our work and make new acquaintances!

Excited to share our new #iclr2023 paper on FaiREE 🧚 We propose a new method to post-process classifiers to achieve finite-sample, distribution-free #fairness guarantees openreview.net/pdf?id=shzu8d6… #trustworthy #AI Great job by Puheng Li and @linjunz_stat!