Assistant Professor @UChicagoCS @DSI_UChicago | PhD @CSDatCMU
Chicago, IL
Joined March 2017
- Tweets184
- Following316
- Followers1.5K
- Likes734
Tian Li retweeted
Excited to share ZEST, an algorithm for sharpness-aware minimization without backpropagation!
ZEST optimizes an exponentially tilted objective that smoothly interpolates between average-loss minimization (t→0) and worst-case SAM (t→♾️).
We derive its zeroth-order gradient, characterize how the sharpness regularizer changes with t, and show that ZEST outperforms other zeroth-order baselines in all experimental settings, sometimes matching or outperforming first-order SAM!
(will be presented at #NeurIPS2026 Sydney 🦘)
🧵
Tian Li retweeted
Excited to share ResearchTrails, a dataset of human research decision trajectories extracted from Git commit histories.
Final papers show the outcome, but they often do not capture the full path: abandoned methods, omitted experiments, and ablations. ResearchTrails recovers these intermediate decisions, and we show that learning from them improves LLMs’ ability to assist with research exploration!
Blog, code, data, and the trajectory visualizer are at researchtrails.org
🧵
check out our work on federating reasoning agents! thread below 👇
🚀Excited to share Federation over Text (FoT), in collaboration with Tahseen Rabbani @TahseenRab74917 , Manzil Zaheer, and Tian Li @litian0331.
Instead of a federation of gradients, we propose a federation of text to improve multi-agent reasoning
Tian Li retweeted
Excited to launch Forecasting Agent Leaderboard
@ProphetArena!
Inaugural agents include Foresight V2 and AG2-Foresight from two fascinating startups @lightningrodai
and @ag2oss -- check them out!
Interested in joining the forecasting competition? Onboarding is simple prophetarena.co/research/age…
🚀 Introducing the Prophet Arena Agent Leaderboard
Prophet Arena is now benchmarking end-to-end forecasting agents.
Previously, we evaluated LLMs forecasting when given a fixed global context. This new leaderboard tests end-to-end forecasting agents that use their own web search, reasoning process and tool use in any creative way they please.
Excited to welcome our first two agents, from two fascinating startups: @lightningrodai and @ag2oss – check them out!
🤖 Add your agent? Just provide an OpenAI-compatible endpoint and we’ll start benchmarking automatically:
👉 prophetarena.co/onboarding
📜 Rules & guidelines:
👉 prophetarena.co/research/age…
Poster #707. My student Xuchen Gong is presenting our #NeurIPS2025 paper on using side data to augment zeroth-order gradient estimates, with applications in differentially private learning! Come chat with us now!
paper link: arxiv.org/abs/2511.10859
Poster #713 happening now (until 2pm)! #NeurIPS2025 Come check out our work on efficient federated adaptive optimization. arxiv.org/abs/2410.18117
Tian Li retweeted
📢 My lab at UCLA is hiring PhD students and postdocs!
Please apply to @CS_UCLA or @CompMedUCLA if you are interested in foundation models and (Gen)AI for health / medicine / science.
More info: cs.ucla.edu/~yuzhe. Please retweet!
Tian Li retweeted
The #MLSys2026 submission deadline is only 2 weeks away (Oct 30)! Submit your best work on ML systems — spanning hardware, compilers, software, models, agents, and eval. This year features both Research and Industry Tracks! Join us in Seattle next spring! mlsys.org
Tian Li retweeted
Job Opportunities Alert
Associate Professor of Data Science: apply.interfolio.com/174768
Assistant Professor of Data Science: apply.interfolio.com/174766
EOE/Vet/Disability
Tian Li retweeted
Come check out what our summer students have been up to! Always a super fun day.
Our summer program participants’ research is well underway, and we can’t wait to show you what they’ve been working on at the DSI Summer Programs 2025: Research Symposium!
Join us Friday, August 8th! datascience.uchicago.edu/eve…
Tian Li retweeted
Random but, I think the best pitch I received for a result was for xitter.cf/litian0331/status/1931…, pitched by Su Hyeong Lee.
Tian Li retweeted
Happening now.
Come tilt your loss at poster w-907
w/ @gh_aminian and @litian0331
[Tue Jul 15]
@gh_aminian & @litian0331 present theoretical results on generalization & robustness of titled empirical risk minimizantion (TERM). We had previously proposed TERM as a simple technique to explore fairness and robustness in ML applications.
xitter.cf/litian0331/status/1372…
Tian Li retweeted
[Tue Jul 15]
@gh_aminian & @litian0331 present theoretical results on generalization & robustness of titled empirical risk minimizantion (TERM). We had previously proposed TERM as a simple technique to explore fairness and robustness in ML applications.
xitter.cf/litian0331/status/1372…
[#ICLR2021 Camera-Ready]
We propose Tilted Empirical Risk Minimization (TERM), a general and flexible extension to ERM framework. TERM increases or decreases the influence of outliers, by tuning a tilt hyperparameter ‘t’ to enable fairness or robustness:
openreview.net/forum?id=K5Ya…
Tian Li retweeted
🔔 Announcing our paper on Natural Language Outlines for Code!
Our vision 🔮 - NL Outlines empower human developers with new forms of AI assistance throughout the software development process 🚀
Paper: arxiv.org/abs/2408.04820
FSE'25 presentation: youtube.com/watch?v=54v7Zphl…
🧵👇
Tian Li retweeted
Excited to share what I worked on during my time at Meta.
- We introduce a Triton-accelerated Transformer with *2-simplicial attention*—a tri-linear generalization of dot-product attention
- We show how to adapt RoPE to tri-linear forms
- We show 2-simplicial attention scales better under token constraints than dot product attention
It was fun collaborating with amazing folks including @dvsaisurya @_arohan_ and others
Tian Li retweeted
Citations are crucial for improving the trustworthiness of LLM outputs. But can we train LLMs to cite their pretraining data *without* retrieval?
New paper from @YukunHuang9 and @sanxing_chen “CitePretrain: Retrieval-Free Knowledge Attribution for LLMs”
arxiv.org/pdf/2506.17585
Tian Li retweeted
Has your LLM been leaking?
We have the perfect solution for you! Introducing our new venture, Gemini Waterproofing (Since 1989)
