Tom Pollard

Where would the field of clinical AI be without PhysioNet? 🎂 This year marks the 25th anniversary of one of the most comprehensive clinical data repositories in existence, best known for hosting datasets like MIMIC and the MIT-BIH Arrhythmia Database. jclinic.mit.edu/physionet-at-2…

Also thank you to everyone at @openminedorg , especially @falcon_ronnie, @lace31692, Dave Buckley, Sameer Wagh, @iamtrask. And thank you to Francis Crawley, @SCEdmunds, Weibin Liu, Yan Li, @geochurch , @tompollard , Juergen Reichart, @ryanjtaft for discussions.

We are underway with our first session: MEDS: Building Models and Tools in a Reproducible Health AI Ecosystem Given by Matthew McDermott! #ML4H2025

PSA: Many of you may have noticed a website floating around called formatmypaper(dot)com. People rightly noted something was fishy. I dig into what happened here: open.substack.com/pub/ubadah/p/b…

Want to continue training an encoder on your own data, but not sure where to start? Our step-by-step guide for reproducing the BioClinical ModernBERT training was just released! 1/5

Exciting to see BioClinical ModernBERT (base) ranked #2 among trending fill-mask models - right after BERT! The large version is currently at #4. Grateful for the interest, and can’t wait to see what projects people apply it to!

Very excited to share the release of BioClinical ModernBERT! Highlights: - biggest and most diverse biomedical and clinical dataset for an encoder - 8192 context - fastest throughput with a variety of inputs - sota results across several tasks - base and large sizes (1/8)

We are excited to share an initial look at what we're building at Lila Sciences! At Lila, we are weaving together several exciting threads that have emerged in AI over the last several years: Highly-capable large language models, generative models of biomolecules and materials, and lab automation to create the next generation of AI models that can run the scientific method at scale. We have assembled a world-class team of scientific and entrepreneurial leaders at this frontier including Geoffrey von Maltzahn, George Church, Rafael Gómez Bombarelli, Molly Gibson, Kenneth Stanley, John Gregoire, and Jacob Feala! If you're excited about joining a team that is scaling reasoning models for some of the most important problems in life and physical sciences, please reach out! You're colleagues will be some of best ML scientists I have ever met at this intersection. Come work with us and push the frontier of what is possible for science. Link to more below!

Only 10 days left to apply to the AHLI CHIL Doctoral Symposium! Are you a PhD student looking for valuable feedback and mentorship on your work? This is the place for you. Call for papers: chil.ahli.cc/submit/doctora… #CHIL2025

🎉 We're thrilled to announce the general release of three de-identified, longitudinal EHR datasets from Stanford Medicine—now freely available for non-commercial research-use worldwide! 🚀 Read our HAI blog post for more details: hai.stanford.edu/news/advancing… 𝗗𝗮𝘁𝗮𝘀𝗲𝘁 𝗦𝘂𝗺𝗺𝗮𝗿𝗶𝗲𝘀 📊 3 longitudinal EHR datasets 👥 Scale: 25,991 patients, 441,680 visits, and 295M clinical events (median: 4,882 events per patient) ⏳ Timeframe: Patient trajectories from 1997 to 2023 (median: 10 years per patient) 𝗦𝘁𝗮𝗻𝗱𝗮𝗿𝗱𝗶𝘇𝗲𝗱 𝗕𝗲𝗻𝗰𝗵𝗺𝗮𝗿𝗸 𝗧𝗮𝘀𝗸𝘀 🎯 Few-shot Learning 🤖 Multimodal Learning & Time-to-Event Modeling ⌛ Long Context Instruction Following & Temporal Reasoning Thanks to @MichaelWornow, Ethan Steinberg, @Zepeng_Huo, @HennyJieCC , @BediSuhana42170 , @AlyssaUnell, @drnigam @StanfordMed

⚠️ Only 2 weeks left to submit to CHIL! ⚠️ We're seeking cutting-edge papers on machine learning + health, covering topics from generative modeling to health policy, and more! Check the full scope: chil.ahli.cc/submit/call-fo…. Don't miss your chance to contribute to #CHIL2025!

$1 million dollars up for grabs from @NIH for the teams that come up with the best measures of research data sharing, also known as a Data Sharing Index (S-index)! freelancer.com/contest/nih-da… HT @NEIDirector

We’re excited to make this year’s CHIL research roundtables as engaging and valuable as possible. To do that, we’d love your input on discussion topics & your interest in leading a roundtable or volunteering as a notetaker. Help us out here:✍️forms.gle/fYS4YMdXuqRXTZ…

We hope STANDING Together helps everyone across the AI development lifecycle to make thoughtful choices about the way they use data, reducing the risk that biases in datasets feed through to biases in algorithms and downstream patient harm. thelancet.com/journals/landi… (10/

These recommendations are the culmination of nearly 3 years of work by an international group of researchers, healthcare professionals, policy experts, funders, medical device regulators, AI/ML developers, and many more besides. (9/

STANDING Together = STANdards for data Diversity, INclusivity and Generalisability. We have worked with >350 stakeholders from 58 countries to agree a set of recommendations to improve the documentation and use of health datasets. (8/

Key point: there is (probably) no such thing as a perfect dataset! Knowledge of a dataset's limitations is not a negative - it is actually a positive, as steps might then be taken to mitigate any issues. Not knowing ≠ there are no issues... (7/

Those using datasets should carefully appraise the suitability of the dataset for their purpose, and consider how they might mitigate any biases or limitations contained within. (6/

To prevent this happening, it's really important that those creating datasets also supply documentation. This should transparently explain what data they contain, and describe any limitations or related issues which those using data should be aware of. (5/