Sam Finlayson

@IAmSamFin

MD-PhD Candidate, working on machine learning for (bio)medicine at + . Past: (BA, MS)

Boston, MA
Vrijeme pridruživanja: listopad 2014.

Medijski sadržaj

  1. 2. velj
    Odgovor korisniku/ci
  2. 2. velj
    Odgovor korisniku/ci
  3. 2. velj
    Odgovor korisniku/ci
  4. 30. sij
    Odgovor korisniku/ci

    And we’re back to normal pre-Hector weather...feel free to visit anytime!

  5. 23. sij

    Live view of me trying to outline the intro for my staple thesis:

  6. 14. sij
    Odgovor korisnicima

    Congrats! Great work and great thread. I look forward to POPCORN and SODA’s follow-up paper...JR MINT? (Would be very refreshing)

  7. 12. sij
    Odgovor korisnicima

    I completely agree with everything you said, and am obviously not a dermatologist, but I have to note potential for what SMBC calls an “exigology” (statement explained by its converse): “Everyone freaks out and over-prepares for medical exams, but almost no one ever fails them”

  8. 11. sij
    Odgovor korisnicima

    Trying to engage with this topic on twitter is akin to getting involved in a land war in Asia.

  9. 10. sij

    I can study one medical fun fact 100 times, diagram it out, turn it into a song, explain it four different ways, etc etc and it still gets cleared 72 hours later. Then I read one weird tidbit about a comic book on Wikipedia and it’s locked in my brain for life!

  10. 9. sij
    Odgovor korisnicima i sljedećem broju korisnika:

    What on earth...

  11. 9. sij
    Odgovor korisniku/ci
  12. 8. sij

    ML world is inconsistent w.r.t. comparing models in papers Things I see: a) No stats at all (just best performance) b) True cross-validation (refit models each time) c) Train-val to fit, then *freeze model* + do bootstrap CIs *within held out set* (e.g )

    Prikaži ovu nit
  13. 6. sij
    Odgovor korisnicima

    I was thinking of Section 7 of the paper which discusses sample-wise non-monotonicity, though admittedly not in a ton of depth .

  14. 30. pro 2019.

    However, arguing that an unproven tech will likely *drop costs* requires magical thinking and willful ignorance of history. (Insult to injury is when such pitches unironically lead off with a graph like below, as if it isn’t evidence *against* new tech til proven otherwise) 3/3

    Prikaži ovu nit
  15. 22. pro 2019.
    Odgovor korisniku/ci

    IDK about this, but I AM certain that this reference, originally from philosophy, is timely in both fields:

  16. 22. pro 2019.
    Odgovor korisniku/ci

    The only problem with this is that Hardy was completely wrong — his work and that of other pure mathematicians have proved extremely useful in practical applications! From Video here (and an enjoyable watch):

  17. 20. pro 2019.

    Submit your best work on integrating ML into healthcare! 500 word abstracts, can be new work or recently published. Fun conversations in a cool setting. "We'll search for tomorrow on every shore And try, oh Lord we'll try, to carry on"

  18. 6. pro 2019.
  19. 2. pro 2019.

    Pete looking good today as he recounts stories from the leading edge of ML+Medicine. Fun stories about: - Guardian Angel project in 90s, work anticipating most of today’s big tech bets (and failures) in med - Work on bioterrorism surveillance presented in DC on Sept *10*, 2001

  20. 29. stu 2019.
    Odgovor korisniku/ci

    At , he/colleagues appear to be raising oodles of $$ on assertion that they’re doing just this. I just wish there were some way to assess the claims, read any tangible details about approach, etc. Else, just sounds (ironically) like empty hype :(((

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:

    ·