What drives innovation in machine learning? In a new paper, we argue that application-driven work is systemically under-valued in the machine learning community, but that it's essential for both innovation and impact. arxiv.org/abs/2403.17381 1/3

📢 Calling all med students and residents I curious about applied AI in healthcare — come join us for case-based discussions and learning from multidisciplinary speakers! We hope it enables more future physicians to co-create or navigate the tools/experiences we might use soon!

.Raj Movva, Pang Wei Koh, and I write for Nature Medicine on using unlabeled data to improve generalization + fairness of medical AI models: nature.com/articles/s4159… We highlight two nice recent papers illustrating this - nature.com/articles/s4159…, nature.com/articles/s4159….

Excited this is out. We tried to understand why an AI tool predicting depression with smartphone data was less accurate in a more heterogenous sample. No surprise - there were different, often conflicting smartphone behaviors predictive of depression across the sample. (1/3)

In a new paper w/ Jon Kleinberg and Hoda Heidari, we explore the incentives and interactions behind the development of AI models. "FINE-TUNING GAMES" I'm in Singapore to present the paper at WWW The Web Conference. It's now published (👇) & was selected for an oral presentation! 🧵

New paper in FAccT 2024 with Aaron Schein, Yixin Wang Nikhil Garg! Many discrimination auditing and electoral tasks use ML to predict race/ethnicity – by discretizing continuous scores. Can the discretization process cause bias in labels and downstream tasks? Yes!

It’s difficult to measure how well LLMs align with human perceptions of safety: Which humans? What constitutes a “perception of safety”? Our take on this question considers *annotation alignment*: the extent to which LLM annotations of safety agree with human annotations. 1/5

This page of common pytorch mistakes is pretty invaluable uvadlc-notebooks.readthedocs.io/en/latest/tuto…

nice writeup about our recent work on underreporting by Cornell Chronicle !

I'm at RL_Conference this week; let me know if you want to meet up! Especially down to chat about RL for building energy management. Bonus: I can intro you to 4 wonderful Olin College undergrads I've worked with this year who are here (and would make excellent PhD Students)✨

Please welcome Deepti Raghavan, joining us this fall as assistant professor! A recent graduate of Stanford University, her research interests focus on operating systems, networking, and machine learning systems: cs.brown.edu/news/2024/08/1…

More training data isn’t always better! We (joint work w.Deb Raji and Irene Chen) will present our new work: The Data Addition Dilemma, at MLHC 2024 this week! (1/5)

There are good arguments for removing race from medical algorithms, but there may be unintended consequences. Our PNAS paper finds that race-blind algorithms can *worsen* racial inequalities, bc they can't adjust for racial disparities in data quality. shorturl.at/7ugW5

🎉📢 I'm on the tenure track job market! 👋✨ I study how people building “AI” think about the downstream harms their work makes possible, and the epistemic and political economic systems which shape their thoughts, and our future.

I’m looking for Winter 2025 (Jan start) internships! I have expertise in multimodal AI, genAI for model explainability, and clinician-AI interaction. Also working on foundation models for health. Please reach out if you think there’s a good fit!