Matt Gardner

@nlpmattg

Senior research scientist at . Original architect of . Co-host of .

Vrijeme pridruživanja: svibanj 2017.

Tweetovi

Blokirali ste korisnika/cu @nlpmattg

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

  1. proslijedio/la je Tweet
    prije 17 sati

    Just a reminder to everyone -- AKBC abstract deadline is only 2 days away! Please submit your papers to this new and exciting conference! (cc )

    Poništi
  2. proslijedio/la je Tweet
    prije 20 sati

    We've released our code here: Hope it helps your research. Demo coming soon!

    Poništi
  3. proslijedio/la je Tweet
    3. velj

    Thanks for having us and the interesting discussion! We were honored to be guests on arguably the best ML/NLP podcast out there!

    Poništi
  4. 3. velj

    104: and talk to us about model distillation, when you try to approximate a large model's decision boundary with a smaller model. After talking about the general area, we dive into DistilBERT.

    Poništi
  5. proslijedio/la je Tweet
    3. velj

    Check out BREAK - a new NLU benchmark for testing the ability of models to break down a question into the required steps for computing its answer. A work by Tomer Wolfson, accepted to TACL 2020.

    Poništi
  6. proslijedio/la je Tweet
    3. velj

    New TACL paper involving a lot of hard work from my twitter-less student Tomer, along with great collab. at AI2 and TAU. Paper/website at 1/2

    Prikaži ovu nit
    Poništi
  7. 3. velj

    And the reason we should probably be kinder to ourselves when we relive these things: I don't remember this. I remember talking to about the poster, but not this particular part.

    Poništi
  8. proslijedio/la je Tweet
    31. sij

    AKBC 2020 Abstract Deadline (February 6) is next week! Full papers are due Feb 13th! Please submit your finest work to the conference. Website () has been updated with submission details.

    Poništi
  9. proslijedio/la je Tweet
    27. sij

    In episode 103, Brendan O’Connor tells about analyzing social media text to address a variety of social and sociolinguistic questions.

    Poništi
  10. proslijedio/la je Tweet
    23. sij

    We're proud to announce the recipients of the AI2 Outstanding Intern of the Year award for 2019: Sarthak Jain and ! Learn more about our Outstanding Intern award and these two fantastic awardees here:

    Poništi
  11. proslijedio/la je Tweet
    22. sij
    Odgovor korisnicima

    So the question, the reason I posted this on twitter, is: at what point is it reasonable as a reviewer to refuse to spend time on a review? And if you get to that point, how do you convey that in a reasonable way in your review?

    Poništi
  12. 22. sij

    This is blatantly dishonest and a waste of everyone's time, particularly that of honest reviewers who want to do a good job and give each paper the review they hope their own paper would get.

    Prikaži ovu nit
    Poništi
  13. 22. sij

    Either of these two approaches is ok. The first one is more likely to result in an accepted paper, realistically, but it's also potentially a whole lot more work. What you *can't* do is just ignore the new work and pretend that your method is SOTA when you know it isn't.

    Prikaži ovu nit
    Poništi
  14. 22. sij

    Or (2) discuss the contributions of the other work, talk about how they relate to your work, give some context to the reader about where your work fits in. Make an argument that you're doing science, not leaderboard chasing, and that your experiments still let us do science.

    Prikaži ovu nit
    Poništi
  15. 22. sij

    Yes, you have to do some work to update your paper, but you have two avenues to proceed: (1) incorporate the basic technique that beat your approach, and show that your contributions are still useful, even on top of whatever else the other work did.

    Prikaži ovu nit
    Poništi
  16. 22. sij

    Secondly, what should an author do in this situation? It really sucks to have a paper rejected in the fast-paced world of leaderboard chasing, I know. Hopefully there was some science in your paper, though, not just a SOTA claim, and that science is likely still valid.

    Prikaži ovu nit
    Poništi
  17. 22. sij

    Asking now, because there's still time for me to fix this review, if I'm missing something here. I very much don't want to make a false accusation, or do the wrong thing. But modern leaderboards leave a paper trail that makes this case pretty obvious.

    Prikaži ovu nit
    Poništi
  18. 22. sij

    So, what should you do as a reviewer in this situation? Do we have any clear guidelines? I basically wrote a strongly worded review that it sure looked like something was up, making me not want to waste my time giving the paper a thorough review. Was that out of line?

    Prikaži ovu nit
    Poništi
  19. 22. sij

    Do not just resubmit your paper without updates to the next conference! *Especially*, do not still claim SOTA when you know full well that you are no longer SOTA. This is a very clear violation of academic integrity.

    Prikaži ovu nit
    Poništi
  20. 22. sij

    So, clearly the authors knew about the better results (they had to have looked at the leaderboard to make their updated submission), but completely omitted them from their paper, even though some were published at EMNLP.

    Prikaži ovu nit
    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:

    ·