Rain and the Sound of Trains

@narrow_mark

We are as gods and might as well get good at it. Machine learning, human enhancement, space enthusiast. He/him.

Great Lakes, United States
Vrijeme pridruživanja: veljača 2015.

Tweetovi

Blokirali ste korisnika/cu @narrow_mark

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

  1. TFW you need to do the writeup for the finished parts of the project.

    Poništi
  2. proslijedio/la je Tweet

    If an algorithm has "mastered natural language", I would expect it to be able to do some of the things language is for -- communicating information, receiving information, acting on the world... Not merely output something that statistically sounds like language.

    Prikaži ovu nit
    Poništi
  3. proslijedio/la je Tweet

    … would now, while everyone's praising us for being a valuable resource to the plural community, be a good time to mention that we don't know how we're going to pay rent next month?

    Prikaži ovu nit
    Poništi
  4. Holy crow, writing is hard.

    Poništi
  5. proslijedio/la je Tweet
    31. sij

    . has a new trailer! Check it out: via

    Poništi
  6. This reminds that Animatrix short World Record, where the narrator says there are frankly oddball ways of discovering the Matrix. Now I'm just looking at my own skin and wondering how I didn't notice this before. Also, I bet Coelary is going to be great, so follow this joker!

    Poništi
  7. Alright, possible lesson learned: if I wake up in the middle of the night and can't get back down after 10 minutes, I should just start working. This time seems to count against my creative/productive capacity for the day, and I am really feeling it right now.

    Poništi
  8. proslijedio/la je Tweet
    31. sij

    An Opinionated Guide to ML Research: “To make breakthroughs with idea-driven research, you need to develop an exceptionally deep understanding of your subject, and a perspective that diverges from the rest of the community—some can do it, but it’s hard.”

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

    Friday (theoretical) physics fun: is spacetime discrete or not? Normally we assume space is infinitely divisible just like the real numbers: between two points there will always be at least one more point. It is not obvious this has to be true. Matter turned out to be atoms.

    Prikaži ovu nit
    Poništi
  10. proslijedio/la je Tweet

    Um, NASA never told me I was going to the moon. Deke told me that I was joining Neil and Buzz on a flight that might be the first to land on the moon. And then we went to work.

    Poništi
  11. proslijedio/la je Tweet
    30. sij

    Good poets are misers, publishing only their very best work, because they are jealous of their reputations. Great poets are magnanimous; they publish their worst along side their best, because there is no damage their mediocrity can do to their heights.

    Poništi
  12. proslijedio/la je Tweet
    30. sij

    When you ask your machine learning expert friend to explain to you what actually is

    Poništi
  13. Great! I've been looking for an excuse to post this old kinetic typography video of Stephen Fry summing up my thoughts on the evolution of language.

    Poništi
  14. Somebody tell me if they think I'm off base, but this is getting suspicious. This model was trained on an unbalanced dataset of ~1100 images, then used on a dataset about half that size. These results are too good. Is there a leak somewhere?

    Poništi
  15. Did something stupid for the sake of demonstration, learning, all that good stuff. Silly me, I was expecting these losses to be backward. You know, like everything on the Internet says overfitting should look like!

    Poništi
  16. Hit that special wall where you realize you started working with filenames like 98cb16e.jpg, and now have to merge three folders filled with filenames like 0000001.jpg. Ideas are fun because you don't hear the cruel, mocking laughter of implementation until it's too late.

    Poništi
  17. proslijedio/la je Tweet
    28. sij

    Over a million particles running real-time on the gpu. They’re attracted to each other while having a weaker desire to reassemble the image.

    Poništi
  18. proslijedio/la je Tweet
    30. sij

    I often struggle to explain that: 1. Generally, the range of things you could possibly do that are worse than nothing is larger than the range of things that are better AND 2. Trying most things is cheap & when trying in earnest, humans are good filterers of possibility space

    Prikaži ovu nit
    Poništi
  19. My head: Why aren't we working? Me: We came to a good stopp- My head: Why aren't we working? Me: We covered a lot of- My head: We should be working right now. Me: I just think we can rel- My head: We both know I can do this all night. Me: 😟

    Poništi
  20. I can't tell if these error rates are suspiciously low given the training and validation losses, or if that's what you would expect from a dataset several times the size of what I'm used to working with.

    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:

    ·