AI prof/Endowed Chair @JohnsHopkins, Foundr @BayesianHealth, investor 15+ AI startups, ex-Stanford AI, MIT TR35, Sloan, @wef YGL, @darpa AI safety & Health AI
Jan 27, 2023 • 9 tweets • 2 min read
AI excitement is back and at an all time high!
Having seen a few cycles of ups & downs with AI over the last 15 years, a few takeaways for investors and those newer to the field. 🧵-
1/ Many ask if all this recent progress is new and surprising.
Exciting - yes!
Overnight success - no!
AI’s been making very consistent progress over the last decade, perhaps two.
Nov 30, 2022 • 9 tweets • 4 min read
Machine learning predictive uncertainty estimates are often unreliable—data shift makes things worse!
How can you audit the uncertainty of an ML prediction, even with biased data?
Very nice work by @kdpsinghlab & team on evaluating Epic's widely deployed sepsis tool. See thread below on issues to consider when evaluating offline model performance and also common ways in which previous studies have erred.
Also some other really key points: model accuracy and lead time (a key metric for improving sepsis outcomes) has to be computed with model in the background.
This point commonly throws off practitioners who're used to seeing interventional studies.
Have been on the phone with friends and family all over India. They went from not caring and celebrating Holi (the festival of colors) to a lock down in 1 week! The worrisome things I learnt:
1/ One of the slums in Mumbai now has COVID-19; these are very densely packed areas with little public health infrastructure. The challenge is there is no reliable reporting of case loads or deaths in poor areas.
Jan 7, 2019 • 23 tweets • 7 min read
Just resurfacing frm 2wks off over the holidays. Visited South Africa—saw gorgeous 🦒🐘🦁s, birds, seaside towns, & spent time thinking! Highlights from 2018(warning: long thread). It's been a hard yr so both bragging & celebrating. Looking fwd to all that’s in store for 2019! 1/
First, on the research front, we made headway in multiple open directions.
As ML is being deployed in domains like healthcare, education, & recruiting, it’s critical we understand scenarios in which model outputs may be unreliable during deployment. 2/
