Suchi Saria

@suchisaria

John C. Malone Prof@ Hopkins, ex-Stanford SAIL, TR35, Sloan, YGL , DARPA YFA Interests: + Augmented Intelligence, Bayes, , Causal Inf, healthcare

Manhattan, NY
Vrijeme pridruživanja: svibanj 2010.

Tweetovi

Blokirali ste korisnika/cu @suchisaria

Jeste li sigurni da želite vidjeti te tweetove? Time nećete deblokirati korisnika/cu @suchisaria

  1. Prikvačeni tweet
    9. lis 2018.

    Seeking a postdoctoral fellow interested in augmented decision-making i.e., exploring principles for building reasoning algorithms that support human experts in decision-making when safety is critical. You will be joining an amazing interdisciplinary team funded by

    Poništi
  2. 2. velj
    Prikaži ovu nit
    Poništi
  3. 2. velj

    Friends, I’m giving a talk at this DC/policy focused meeting: I’m looking for egs of tasteless /ML algorithmic research or poor implementations that could have harmful consequences, to highlight what they did wrong & what right might have looked like.

    Prikaži ovu nit
    Poništi
  4. 12. sij

    My obs: Many DL/CI/AI debates sparked by non-expert twitter provocateurs & core researchers feel the pressure to respond! Our productive scientists ought to be helping make research progress! Cur back & forth presents a hostile image often rehashing the same topic. Am I wrong?

    Prikaži ovu nit
    Poništi
  5. 12. sij

    This study’s caused a stir. My take: -regression to the mean is a common flaw in pre/post studies; this intervention was esp vulnerable b/c it selected patients at the peak. This could have been ~corrected for(even w/o RCT) -but has standard of care also improved sig over time?

    Poništi
  6. 10. sij

    Just noticed our cookbook shelf has one VERY odd resident! (Organized by our movers a year ago.)

    Poništi
  7. 4. sij

    This is no different than how other non-AI diagnostic/screening tests have been introduced before: analyze the safety and efficacy — ie, downstream consequences of over treatment and missed diagnosis before introducing a protocol that’s deployed at scale

    Poništi
  8. 25. pro 2019.

    This isn’t surprising. DL uses a class of models; I think of the rest as ML—learning frm small sample, correcting 4 bias, inferring causal effects... But at the height of DL fever, DL got branded as a separate field & known learning principles got reinvented w/ DNN in the title

    Poništi
  9. 22. pro 2019.

    If you love puzzles and decoding magic, and want to be entertained, visit the crossword constructor ‘s magic show. It is absolutely brilliant! Mind blown. (And, the perfect idea for a NYC date!)

    Poništi
  10. 10. pro 2019.

    Imagine discussing machine learning and medicine in Bermuda with other top researchers across informatics, policy, clinical translation, and ML. That's what this meeting is about! Abstracts due soon. If you have questions, send me a note.

    Poništi
  11. proslijedio/la je Tweet

    Finally, a paper on why "AI for good" is an empty phrase without a theory of change. What is "good" is never articulated in the rush to tech solutions, while alternative reforms are overlooked. Read 's piece before it blows up at

    Poništi
  12. proslijedio/la je Tweet
    3. pro 2019.

    Watch “Panel: How Connected Data Can Power a Learning Healthcare System” on Valuable point by : care providers should see the system improving their productivity, helping them do something better esp when already have lots of things to do

    Poništi
  13. 24. stu 2019.

    Related: Published in Biostatistics this week, this article overviews learning methods that correct for statistical biases commonly observed when developing tools from real-world EHR data.

    Prikaži ovu nit
    Poništi
  14. 24. stu 2019.

    Nice paper by et al. w/ examples disentangling social vs statistical bias & ideas for how (well-designed) AI/ML tools can help mitigate social bias: We can't solve a problem we don't understand. To address bias, we need a cleaner taxonomy.

    Prikaži ovu nit
    Poništi
  15. proslijedio/la je Tweet
    23. stu 2019.

    Why would we use to predict risk of kidney injury based on contrast exposure during stent placement? Because the relationship is complex. Non-linear and nuanced. We can individualize the predictions.

    Poništi
  16. proslijedio/la je Tweet
    19. stu 2019.

    Highly recommend checking out this issue. and I discuss problems w/ differing training and deployment conditions, and how causality gives us a language/tools to express and identify the problem, and tools to develop new robust algorithms.

    Poništi
  17. proslijedio/la je Tweet
    19. stu 2019.

    As the editors of , and I are thrilled to share this free access multidisciplinary collection of commentaries on machine learning for causal inference. All 5 pieces are linked in our editorial about the series:

    , , i još njih 4
    Poništi
  18. proslijedio/la je Tweet
    12. stu 2019.

    Today is a tipping point for value-based care in . and are now part of , our value-based program that holds health systems and jointly accountable for better health outcomes and lower costs.

    , , i još njih 4
    Prikaži ovu nit
    Poništi
  19. proslijedio/la je Tweet
    12. stu 2019.

    Excellent HBR article citing brilliant who I’d cite too (& who’s work I only partly understand) - should consider some of this. Cc:

    Poništi
  20. proslijedio/la je Tweet

    Here's the thing. 1/ The perception of Google culture is that no-one curbs the curiosity of engineers Google can sign a BAA, but they have to convince people that they actually have controls in place to ensure that the data is only being used for the purposes of the agreement

    Prikaži ovu nit
    Poništi
  21. 10. stu 2019.

    Aww, thx! My favorite meetings are when the discussants & speakers prior to me motivate new talk threads. Vincent Liu from led to my creating of this slide asking when do we need high quality models vs when are simple rule-based/epic-style predictive systems enough?

    Poništi

Čini se da učitavanje traje već neko vrijeme.

Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.

    Možda bi vam se svidjelo i ovo:

    ·