Tweetovi

Blokirali ste korisnika/cu @theshawwn

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

  1. Prikvačeni tweet
    31. pro 2018.

    now has a "suggest title" link on the submission page. Submit a story! Thanks to the crew for suggesting this.

    Poništi
  2. prije 3 sata
    Prikaži ovu nit
    Poništi
  3. prije 3 sata

    In section 4 of the biggan paper, they present a technique that seems to predict mode collapse. Anyone know how these graphs were made? Or describe it in pseudocode?

    Prikaži ovu nit
    Poništi
  4. proslijedio/la je Tweet

    i made another one because i like them and also why not 🌲🎧

    Prikaži ovu nit
    Poništi
  5. 2. velj

    That feeling when you want the TPUs to preempt so that you can reset them and go to bed. I should really stop creating TPUs at 1am.

    Poništi
  6. proslijedio/la je Tweet
    2. velj

    i guess it was a matter of time before someone tried feeding tentacle porn to a stylegan

    Poništi
  7. 2. velj

    The “outside ML/AI” caveat made me smile. It’s why I fucking love working in AI. The field is so new! Small ideas from other fields can have huge impacts here. (Making it easy to add beautiful UI for AI projects, for example.)

    Poništi
  8. proslijedio/la je Tweet
    2. velj

    what the fuck does “pc load letter” mean

    Prikaži ovu nit
    Poništi
  9. 2. velj

    I wonder when twitter will mark these as sensitive content. By the way, you can watch it train here:

    Prikaži ovu nit
    Poništi
  10. 2. velj

    Remarkable how much difference a few days makes.

    Prikaži ovu nit
    Poništi
  11. proslijedio/la je Tweet
    1. velj

    random walk looks kinda cool. I might try to get a bigger dataset or play with hyperparams to train it further. This is ~300 kimg

    Prikaži ovu nit
    Poništi
  12. 1. velj

    Started training stylegan2 on tentacle porn last night. The early results are neat (and SFW). I decided to do tentacle porn because it seems challenging for stylegan to learn, and the idea was too funny to pass up. Credit to that one dude on IRC who gathered 2,483 images.

    Prikaži ovu nit
    Poništi
  13. proslijedio/la je Tweet
    1. velj
    Odgovor korisniku/ci
    Poništi
  14. proslijedio/la je Tweet
    31. sij

    Cooperative Communication Networks (CCNs [or whatever we want to call them]) allow us to generate interesting, noise-robust representations with no dataset. See and

    Prikaži ovu nit
    Poništi
  15. proslijedio/la je Tweet
    31. sij

    Reversible 8-bit lava lamp

    Poništi
  16. proslijedio/la je Tweet
    31. sij

    Have you ever wanted to interpolate with categorical variables? What does that even mean? Let me show you!

    Prikaži ovu nit
    Poništi
  17. 1. velj

    🐍”It’s pronounced ‘a sssscalar’.”

    Prikaži ovu nit
    Poništi
  18. 1. velj

    Idea: An optimizer like Adam (one estimate per weight) + LARS (a learning rate per layer). I’m not sure it makes sense, but it seems worth exploring.

    Poništi
  19. 1. velj

    7. Beautiful code. I was surprised just how pretty the benchmark code was. And I think it’s why it’s fast. Wild animals are beautiful because they have hard lives. This code is beautiful because it has zero overhead. Nothing is wasted.

    Prikaži ovu nit
    Poništi
  20. 31. sij

    6. LARS optimizer: For TPUv3-256 and above, the batch size is 32,768. Such huge batch sizes cause problems, because you have to raise the learning rate as you increase the batch size. Normally this causes divergence. LARS has a learning rate per layer.

    Prikaži ovu nit
    Poništi
  21. 31. sij

    They exploit this by decoding the jpeg on the TPU’s CPU, random cropping it, then sending it to an individual TPU core for processing. Input processing happens in parallel with resnet training. The TPU’s CPU prepares the next inputs while each TPU core trains on the prev batch.

    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:

    ·