nick frosst  

@nickfrosst

research engineer Toronto lab : - singer in :

Toronto, Ontario
Vrijeme pridruživanja: travanj 2012.

Medijski sadržaj

  1. 5. sij

    I turned 27 today and am feeling pretty good about it :) I have been celebrating on a mountain in Norway with some favorite people. 2020 is gonna be a good year :)

    9 replies 2 proslijeđena tweeta 106 korisnika označava da im se sviđa
  2. 14. pro 2019.

    My band is presenting a poster at NeurIPS today :) also we will be playing a set at the banquet at 7:30. Come check it and other great artistic applications of machine learning out at the ml creativity workshop :)

    20 proslijeđenih tweetova 163 korisnika označavaju da im se sviđa
  3. 5. pro 2019.

    This actually makes me feel sweet :) I am really excited to get more music out soon :) also side note Spotify has super nice statistics for musicians. I bet there a ton of opportunities for ML there.

    5 proslijeđenih tweetova 82 korisnika označavaju da im se sviđa
  4. 4. pro 2019.

    So my band is coming to ! We will be playing a set at the banquet on Saturday and presenting a paper at the Creativity Workshop! These two worlds don't normally collide like this for me, so i am really excited to merge some ML stuff with music!

    8 proslijeđenih tweetova 114 korisnika označava da im se sviđa
  5. 21. stu 2019.

    I'm gonna tell my kids this was Artificial Intelligence.

    13 proslijeđenih tweetova 110 korisnika označava da im se sviđa
  6. 4. stu 2019.
    Odgovor korisniku/ci

    1 reply 1 korisnik označava da mu se sviđa
  7. 5. lis 2019.

    It's been great working with these past few months! She completed 2 papers over the course of her summer internship and we are very excited to release them!

    Korisnici , i
    5 proslijeđenih tweetova 127 korisnika označava da im se sviđa
  8. 8. srp 2019.

    This is true even with an epsilon of 0.1, as is standard in the literature! Below are randomly sampled successful and unsuccessful targeted R-PGD attacks against a capsule network. Note how many resemble the target class 6/7

    1 reply 1 proslijeđeni tweet 6 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
  9. 8. srp 2019.

    however, the reconstruction error is differentiable; we can make a stronger attack by taking this into account. We call this attack R(econstruction)-PGD. It can trick our detection, but the results are not really adversarial, they resemble members of the target class! 5/7

    1 reply 1 proslijeđeni tweet 8 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
  10. 8. srp 2019.

    With this strategy we are able to accurately detect many types of adversarial attacks on MNIST, fashionMNIST, and SVHN. We are also able to use this mechanism as a general outlier detection strategy 4/7

    1 reply 1 proslijeđeni tweet 6 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
  11. 8. srp 2019.

    We can create a detection algorithm by defining a threshold for reconstruction error, and flag inputs as adversarial if the reconstruction error exceeds this threshold. This strategy can be extended to CNNs as well, though it does not work as well 3/7

    1 reply 1 proslijeđeni tweet 7 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
  12. 8. srp 2019.

    The problem with adversarial examples is that they dont look like what they are classified as. Capsule networks output both a classification and a reconstruction of the input conditioned on the classification. A reconstruction of an adversarial looks different from the input 2/7

    1 reply 1 proslijeđeni tweet 16 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
  13. 2. srp 2019.

    weird flex but ok.

    13 korisnika označava da im se sviđa
  14. 17. lip 2019.

    Update: it's gotten sunnier and crazier

    15 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
  15. 17. lip 2019.

    For a city that has never held a championship parade before, I think we are doing a pretty damn good job.

    47 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
  16. 14. lip 2019.

    Our poster session at ICML went well and the raptors won the championship. Pretty good night if you ask me :)

    48 korisnika označava da im se sviđa
  17. 13. lip 2019.

    If you are at and want to learn about a wierd loss function and all the stuff you can do with it, come to Hall A at 12:05. I will be giving a 5 minute overview of my work with and . Drop by the poster session tonight for more info :)

    9 proslijeđenih tweetova 47 korisnika označava da im se sviđa
  18. 21. svi 2019.

    and I wrote a blog explaining and demoing the DKNN method and how it can be used in conjunction with Soft Nearest Neighbor Loss to detect adversarial examples and outlier data. It even has some little in browser neural networks :)

    5 korisnika označava da im se sviđa
  19. 16. svi 2019.
    Odgovor korisnicima

    The cycle continues.

    1 korisnik označava da mu se sviđa
  20. 10. svi 2019.

    This is my experience in the United States most of the time

    1 reply 11 korisnika označava da im se sviđa

@nickfrosst još nije poslao/la nijedan tweet s fotografijama ili videozapisima.

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:

    ·