Colin Raffel

@colinraffel

Research scientist (formerly resident) at Google Brain. Joining as an assistant professor in fall 2020. Here for the preprints. He/him/his

San Francisco, CA
Vrijeme pridruživanja: ožujak 2017.

Tweetovi

Blokirali ste korisnika/cu @colinraffel

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

  1. Prikvačeni tweet
    12. stu 2019.

    I'm starting a professorship in the CS department at UNC in fall 2020 (!!) and am hiring students! If you're interested in doing a PhD please get in touch. More info here:

    Prikaži ovu nit
    Poništi
  2. proslijedio/la je Tweet
    prije 8 sati
    Poništi
  3. 30. sij

    Thanks for hosting! Always a pleasure!!

    Prikaži ovu nit
    Poništi
  4. 30. sij

    Had a great time talking about T5 and chatting with students yesterday! I'm waiting to put slides online until I finish annotating them; in the meantime here is a recording of the same talk from when I gave it earlier this month:

    Prikaži ovu nit
    Poništi
  5. 23. sij
    Prikaži ovu nit
    Poništi
  6. 23. sij

    Hot take: Mathiness [1] is like an adversarial patch [2] for ML conference reviewers: Mathiness causes a reviewer to classify the paper as "accept" regardless of whether the math is useful/valid and the paper is any good. [3] Fig. 6 has some empirical evidence of this. (refs ⬇️)

    Prikaži ovu nit
    Poništi
  7. proslijedio/la je Tweet
    22. sij

    FixMatch: focusing on simplicity for semi-supervised learning and improving state of the art (CIFAR 94.9% with 250 labels, 88.6% with 40). Collaboration with Kihyuk Sohn, Nicholas Carlini

    Prikaži ovu nit
    Poništi
  8. 17. pro 2019.

    Today, the T5 team competed against T5 in a "pub quiz" on (context-free) questions from the TriviaQA/NQ validation sets. We LOST! We only got 20% right; T5 got 35%. To see how to fine-tune T5 on context-free QA (or any other task) with a free TPU, check out our Colab tutorial ⬇️

    Poništi
  9. proslijedio/la je Tweet
    13. pro 2019.

    I'm so excited about the program we've put together for Saturday's ML for Creativity and Design Workshop 3.0. Aside from the amazing accepted talks and posters, we have a diverse set of invited speakers I want to highlight in this thread.

    Prikaži ovu nit
    Poništi
  10. 12. pro 2019.

    Me at the poster session when I see a paper I reviewed and fought for accepting

    Poništi
  11. 12. pro 2019.

    tips day 5 (h/t )! Conferences are a parade of successes. Remember that for every impressive paper there are many (unpublished) ideas that didn't pan out. Take this opportunity to ask people about negative results!

    Poništi
  12. 11. pro 2019.

    Our code is available here: MixMatch is joint work with , , Nicholas Carlini, , and . Thanks for coming to my Twitter poster presentation! 11/11

    Prikaži ovu nit
    Poništi
  13. 11. pro 2019.

    The MixMatch paper includes experiments on other datasets as well as a nice ablation study. In the past 6 months, our results have been beaten - including by us! Look out for more exciting SSL results soon. 10/11

    Prikaži ovu nit
    Poništi
  14. 11. pro 2019.

    The combination of these ingredients produced SoTA results in the realistic SSL setting () when MixMatch came out. For example, we achieved an error rate of about 11% with only 250 labels on CIFAR-10. 9/11

    Prikaži ovu nit
    Poništi
  15. 11. pro 2019.

    For regularization, we use weight decay and MixUp () across both labeled and unlabeled examples. This is super important to get good performance. 8/11

    Prikaži ovu nit
    Poništi
  16. 11. pro 2019.

    After we've obtained label guesses, we just proceed as if we're doing supervised learning, with label guesses as targets for unlabeled examples and the ground-truth labels as targets for labeled examples. 7/11

    Prikaži ovu nit
    Poništi
  17. 11. pro 2019.

    Sharpening simply corresponds to lowering the distribution's temperature. The sharpened average prediction is then used as the label guess for the original unlabeled image. 6/11

    Prikaži ovu nit
    Poništi
  18. 11. pro 2019.

    MixMatch is a *holistic* SSL algorithm which combines these ingredients in a simple recipe. MixMatch first feeds k augmented versions of an unlabeled image to the model to obtain k predictions. It then averages the k predictions and "sharpens" the result. 5/11

    Prikaži ovu nit
    Poništi
  19. 11. pro 2019.

    Another important ingredient when doing semi-supervised learning is regularization, because typically we only have a few labels and it's easy for the model to memorize them. 4/11

    Prikaži ovu nit
    Poništi
  20. 11. pro 2019.

    Two common ingredients for producing label guesses are consistency regularization ("When I perturb the input or model, the model's prediction shouldn't change.") and entropy minimization ("The model should output low-entropy/confident predictions on unlabeled data.") 3/11

    Prikaži ovu nit
    Poništi
  21. 11. pro 2019.

    The goal in semi-supervised learning (SSL) is to use unlabeled data to improve a model's performance. Many approaches do this by using the model to produce "label guesses" for unlabeled data, and then training the model to predict those guesses. 2/11

    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:

    ·