AI Professor @UCIrvine | Associate Director, AI in Science Institute | Formerly @blei_lab, @Princeton | Chair @aistats_conf 2025 | AI Resident @ChanZuckerberg
A distributional perspective on climate projections: combining observations and simulations through diffusion guidance, and using simulated futures to tune the balance.
Across 17 climate-model experiments, guidance more than halves median biases in predicted temperature extremes.
A lot of work went into this. Congrats to @f_immorlano and the team!
17 climate models, 17 different futures, and 1 observed past that reveals models’ biases. How confident can we be in climate projections?
New preprint: diffusion guidance steers a generative emulator of 17 CMIP6 models toward observations
📑 → https://t.co/7F3si91OPX
🧵 1/4
Robots learn by imitating human teleop demos.
But some tasks are too hard to demonstrate
Raw VR teleop success for hard tasks: 0% 🍅, 10% 🖍️, 0% 🍷
Our CoRL work, GLIDE, can bump this to 90%!
How? By smartly editing teleop code.
A refreshingly robotics-first way to use LLMs!
Big news from our research lab at UC Irvine, and ETH collaborators! 🚀
We introduce Parallel Language Models: up to 5.7× faster inference at autoregressive-level quality, scaling to 3B parameters.
We'll open-source everything! Huge congratulations to the amazing team! 🎉
This will change how LLMs are trained: Parallel Language Models predict several tokens in parallel at autoregressive quality. Add a parallel decoder at pretraining, get 5.7x speedup after posttraining.
🧐 → https://t.co/vXTeHmLOEe
ℹ️ → https://t.co/UuhnIBJZ6x
Read on: 🧵 1/4
I've been advising and working with @lumaril over the past few months, and the company is coming out of stealth today! Applying AI to real-world industrial operations is a fascinating and rewarding area to work in, and it's been great to see the early successes across several industries.
Today we’re launching Lumaril.
We’re an industrial autonomy company helping America’s existing factories run at their true capacity, with more production and less waste.
The US is home to hundreds of thousands of plants, run by senior operators with decades of tacit knowledge. Much of it is unwritten, and a growing wave of retirements is taking it out the door.
If we let these facilities go dark, we don't just lose great companies and the towns that depend on them. We lose the living blueprints for making the goods and materials this country can’t afford to do without.
Reindustrialization won’t come from new factories alone. The plants we already have are great American businesses, built over generations, with enormous potential to grow. We believe helping them scale is the fastest path to restoring industrial capacity.
The last few years have been tough for those applying for faculty positions. I'm happy to announce that my department has an opening in AI and ML.
AI has historically been one of our strengths. Statistics is literally just two floors below, and we have centers and institutes like the AI in Science Institute and the Center for Machine Learning.
Perks include: a great research environment, a politics-free work environment, proximity to the tech hubs in LA and Orange County, ... did I mention the beach?
https://t.co/vbSaTowOH1
Objects in robot demos are awfully patient 🙃
We’ve spent months on tasks where they aren’t. Catching balls, pouring into a moving pot, taking a bottle from someone walking by.
Two additions helped π0.5, RTC, & VLASH handle dynamic manipulation! More from our CoRL'26 oral in 🧵
Daniel & Jimin ran 1,500+ physical robot evaluations to put it to the test, with 50 runs per reported success rate.
As always, we’re making everything public: https://t.co/3IeMVCJSXQ
Diffusion models can do more than generate images: they can save us from transmitting all their detail.
Our IEEE BITS article reviews generative compression and near-optimal rate–distortion–realism tradeoffs. Led by @YiboYang, who did the lion’s share.
https://t.co/DSurFJo2IW
Our paper "Trading off Controllability and Realism in Generative Physics Emulators" received an Outstanding Paper Award at the AI4Physics Workshop at #ICML 2026. Congrats to my co-authors, especially lead authors @f_immorlano and Elijah Tavares!
AI for science needs calibrated models and uncertainty, but we show standard test-time diffusion guidance methods miss the Bayesian posterior. Our Calibrated Bayesian Guidance (CBG) fixes this, enabling Bayesian inference and setting a new SOTA for black-hole imaging.
Promoted from Associate to Full Professor. Fewer letters, more committees. 🙂 Thanks to all the graduate students, postdocs, and colleagues. I wouldn't have made it here without you!
If I had to bet on LLM infrastructure impact, this might become the most important paper from my lab. This ICLR paper intrinsically parallelizes language models, building on ideas from normalizing flows. 🚀
Excited to see seven years of work come together: we can now consistently predict thermodynamic properties of fluid mixtures directly from molecular structure using machine learning. 🧪🤖
Our latest work just got accepted at Nature Communications: https://t.co/UcvrKvhbjo (1/4)
Our latest model takes this to a new level: predicting excess Gibbs energies of mixtures from the SMILES representations of their constituents. This enables predictions across temperatures and compositions, central to chemical process and solvent design. (3/4)
We've known that diffusion models are theoretically very good lossy data compressors , but how can we actually implement this idea in practice? I discuss this and related topics in a new review article on diffusion-based generative compression https://t.co/IEhezuxTB0
ProbML 2026 (formerly AABI) invites submissions on probabilistic ML (Bayesian & beyond), July 5 in Seoul (co-located with ICML). Website: https://t.co/ttTacxNCVn. Tracks: proceedings (PMLR), workshop, fast track. New focus includes healthcare & climate!
Submit by: 20 March 2026