Yuzhe Yang

@yuzheyang.bsky.social

AI Prof @ UCLA | RS @ Google | PhD @ MIT #ML, #AI, #health https://www.cs.ucla.edu/~yuzhe https://yang-ai-lab.github.io/

๐Ÿ“ก Introducing ๐—ฆ๐—ฒ๐—ป๐˜€๐—ผ๐—ฟ-๐—Ÿ๐—ฎ๐—ป๐—ด๐˜‚๐—ฎ๐—ด๐—ฒ-๐—”๐—ฐ๐˜๐—ถ๐—ผ๐—ป (๐—ฆ๐—Ÿ๐—”) models โ€” a new class of models that connect sensor time series, natural language, and actions in one unified framework. ๐Ÿ‘‡ ๐ŸŒ: yang-ai-lab.github.io/OpenSLA ๐Ÿ“„: arxiv.org/abs/2607.06617 ๐Ÿ’ป: github.com/yang-ai-lab/... ๐Ÿงต

โŒš ๐— ๐—ผ๐˜๐—ถ๐—ผ๐—ป ๐—ถ๐˜€ ๐˜‚๐—ป๐—ถ๐˜ƒ๐—ฒ๐—ฟ๐˜€๐—ฎ๐—น. ๐—ง๐—ต๐—ฒ ๐—บ๐—ผ๐—ฑ๐—ฒ๐—น๐˜€ ๐—ฏ๐˜‚๐—ถ๐—น๐˜ ๐—ณ๐—ผ๐—ฟ ๐—ถ๐˜ ๐—ฎ๐—ฟ๐—ฒ ๐—ป๐—ผ๐˜. Introducing ๐—œ๐—ป๐—ฒ๐—ฟ๐˜๐—ถ๐—ฎ-๐Ÿญ โ€” a fully open cookbook for building unified wearable motion foundation models. ๐Ÿ‘‡ ๐ŸŒ: yang-ai-lab.github.io/Inertia-1 ๐Ÿ“„: arxiv.org/abs/2607.06617 ๐Ÿ’ป: github.com/yang-ai-lab/... ๐Ÿงต

Generative models have transformed language and vision. But the physical world runs on something else: ๐™จ๐™š๐™ฃ๐™จ๐™ค๐™ง ๐™ฉ๐™ž๐™ข๐™š ๐™จ๐™š๐™ง๐™ž๐™š๐™จ. Today, we are excited to introduce ๐—ฆ๐—ฒ๐—ป๐˜€๐—ผ๐—ฟ๐—š๐—ฒ๐—ป โ€” fully open, large-scale study of generative modeling for real-world sensor time series. ๐ŸŒ: yang-ai-lab.github.io/sensor-gen

I will be at #ICML in Seoul next week! Our group will present 3 papers there: ๐Ÿ“ˆ ๐—›๐—˜๐—”๐—ฅ๐—ง๐—ฆ โ€” Jul 7, 2 PM, Hall A #712 ๐ŸŒ™ ๐—ฆ๐—น๐—ฒ๐—ฒ๐—ฝ๐—Ÿ๐—  โ€” Jul 8, 5 PM, Hall A #4601 ๐Ÿ›Œ ๐—ข๐—ฝ๐—ฒ๐—ป ๐—ฆ๐—น๐—ฒ๐—ฒ๐—ฝ ๐—™๐—  (๐—ข๐—ฆ๐—™) โ€” Jul 9, 10:30 AM, Hall A #304 I'd be happy to meet and chat! DMs are open. Come say hello! ๐Ÿ‘‹

Bild

โค๏ธ ๐—›๐—˜๐—”๐—ฅ๐—ง๐—ฆ keeps growing with a new open-source release! yang-ai-lab.github.io/HEARTS Our goal with ๐—›๐—˜๐—”๐—ฅ๐—ง๐—ฆ has always been to build a living benchmark for health time-series reasoning. This second wave makes it easier to use, cheaper to evaluate, and more open for the community.

Bild

๐Ÿšง What if your regression model is learning shortcuts too? Introducing Deep Spurious Regression (DSR) โ€” our new framework for studying spurious correlations in continuous prediction. ๐Ÿงต๐Ÿ‘‡

๐Ÿš€ The website for the ๐—›๐—ฒ๐—ฎ๐—น๐˜๐—ต ๐—œ๐—ป๐˜๐—ฒ๐—น๐—น๐—ถ๐—ด๐—ฒ๐—ป๐—ฐ๐—ฒ ๐—Ÿ๐—ฎ๐—ฏ (๐—›๐—”๐—œ๐—Ÿ) at UCLA is (finally) live! Check it out โ†’ yang-ai-lab.github.io ๐Ÿ“ข We are hiring! ๐—›๐—”๐—œ๐—Ÿ has multiple openings for postdocs and is also actively recruiting at all levels. Please visit our website for application and contact details! #AI #Hiring

Health Intelligence Lab - UCLA

Advancing scalable, multisensory, and generative AI for healthcare, medicine, and scientific discovery.

yang-ai-lab.github.io

Exciting news for the Health Intelligence Lab at UCLA: 3/3 papers accepted to #ICML (1 as Spotlight) ๐ŸŽ‰ ๐—ฆ๐—น๐—ฒ๐—ฒ๐—ฝ๐—Ÿ๐— : ๐—ก๐—ฎ๐˜๐˜‚๐—ฟ๐—ฎ๐—น-๐—Ÿ๐—ฎ๐—ป๐—ด๐˜‚๐—ฎ๐—ด๐—ฒ ๐—œ๐—ป๐˜๐—ฒ๐—น๐—น๐—ถ๐—ด๐—ฒ๐—ป๐—ฐ๐—ฒ ๐—ณ๐—ผ๐—ฟ ๐—›๐˜‚๐—บ๐—ฎ๐—ป ๐—ฆ๐—น๐—ฒ๐—ฒ๐—ฝ (spotlight) ๐—ข๐—ฆ๐—™: ๐—ข๐—ป ๐—ฃ๐—ฟ๐—ฒ-๐˜๐—ฟ๐—ฎ๐—ถ๐—ป๐—ถ๐—ป๐—ด ๐—ฎ๐—ป๐—ฑ ๐—ฆ๐—ฐ๐—ฎ๐—น๐—ถ๐—ป๐—ด ๐—ผ๐—ณ ๐—ฆ๐—น๐—ฒ๐—ฒ๐—ฝ ๐—™๐—ผ๐˜‚๐—ป๐—ฑ๐—ฎ๐˜๐—ถ๐—ผ๐—ป ๐— ๐—ผ๐—ฑ๐—ฒ๐—น๐˜€ ๐—›๐—˜๐—”๐—ฅ๐—ง๐—ฆ: ๐—•๐—ฒ๐—ป๐—ฐ๐—ต๐—บ๐—ฎ๐—ฟ๐—ธ๐—ถ๐—ป๐—ด ๐—Ÿ๐—Ÿ๐—  ๐—ฅ๐—ฒ๐—ฎ๐˜€๐—ผ๐—ป๐—ถ๐—ป๐—ด ๐—ผ๐—ป ๐—›๐—ฒ๐—ฎ๐—น๐˜๐—ต ๐—ง๐—ถ๐—บ๐—ฒ ๐—ฆ๐—ฒ๐—ฟ๐—ถ๐—ฒ๐˜€ ๐Ÿ‘‡

BildBildBildBild

๐Ÿšจ Let your wearable data "speak" for themselves! โŒš๏ธ๐Ÿ—ฃ๏ธ Introducing *SensorLM*, a family of sensor-language foundation models, trained on ~60 million hours of data from >103K people, enabling robust wearable sensor data understanding with natural language. ๐Ÿงต

Bild

๐Ÿฉปโš–๏ธ AI underdiagnoses Black female patients A new study found that expert-level vision-language models for chest X-rays systematically underdiagnose marginalised groups โ€“ especially Black women โ€“ more than radiologists. ๐Ÿ”— doi.org/10.1126/sciadv.adq0305 #SciComm #AI #HealthEquity ๐Ÿงช

Demographic bias of expert-level vision-language foundation models in medical imaging

Compared to certified radiologists, expert-level AI models show notable and consistent demographic biases across pathologies.

doi.org

Do foundation models in medical imaging see everyone fairly?๐Ÿค” Excited to share our new Science Advances paper uncovering & auditing demographic biases of expert-level VLMs, and comparing to board-certified radiologists๐Ÿง‘โ€โš•๏ธ ๐Ÿ“„science.org/doi/10.1126/sciadv.adq0305 ๐Ÿ’ปgithub.com/YyzHarry/vlm-fairness (1/)

Bild

Just published in Nature Biomedical Engineering! Working with the incredible PhD student Wei Qiu and our brilliant collaborator Kamila Naxerova at Harvard was a great pleasure. Our deep profiling framework enables us to view 18 human cancers through the lens of AI! www.nature.com/articles/s41...

Deep profiling of gene expression across 18 human cancers - Nature Biomedical Engineering

Using unsupervised deep learning to generate low-dimensional latent spaces for gene-expression data can unveil biological insight across cancers.

nature.com

Seven years ago, Scott Lundberg, presented our SHAP framework at the NeurIPS 2017 conference. Since then, SHAP has become one of the most widely used feature attribution methods, with our paper receiving approximately 30,000 citations. It's wonderful that SHAP's birthday aligns perfectly with mine!๐Ÿ˜Š

Bild