Roman Ring

@Inoryy

MSc in CS student ; RE intern alum ; AI, Deep Reinforcement Learning, Computational Neuroscience

Estonia
inoryy.com
Vrijeme pridruživanja: srpanj 2011.
47 fotografija ili videozapisa Fotografije i videozapisi

Tweetovi

Blokirali ste korisnika/cu @Inoryy

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

  1. Prikvačeni tweet
    26. sij

    From passing words of encouragement to years of mentoring, I was incredibly lucky to have so much support in life. It's time to pay it forward and I'm committing to never turn down anyone seeking guidance on anything. From coding DRL agents to playing games, I am here to help.

    1 reply 3 proslijeđena tweeta 16 korisnika označava da im se sviđa
    Poništi
  2. 29. sij

    Had to do a double take when I realized featured my blog post in his latest video, what a nice start of the day :)

    1 reply 1 proslijeđeni tweet 8 korisnika označava da im se sviđa
    Poništi
  3. 29. sij

    Quite an interesting take on higher level ML framework architecture. Supports JAX as a backend too!

    Introducing the new Thinc, a refreshing functional take on deep learning! 🔮 Static type checking 🔥 Mix , & ⛓️ Integrated config system 🧮 Extensible backends incl. JAX (experimental) 🧬 Variable-length sequences & more
    Prikaži ovu nit
    3 korisnika označavaju da im se sviđa
    Poništi
  4. 26. sij

    Disappointed in continued support for this fake controversy. Nobody claims current DL is the solution, it's just one stepping stone. GPT-2 was overhyped but a good step forward. So many people tried explaining it by now that it's clear author is acting in bad faith.

    GPT-2 can generate some impressive looking sentences, but claims that "anything that looks like genuine understanding is just an illusion."
    1 proslijeđeni tweet 3 korisnika označavaju da im se sviđa
    Poništi
  5. 23. sij

    Julia + Zygote / Flux is another exciting ML technology worth exploring and following.

    Odgovor korisnicima @MikeJInnes @Inoryy i sljedećem broju korisnika: 2
    Julia's Python interop (PyCall) is not experimental at all: it's widely used, mature, & has existed since 2011. Classic 🤯demo is recursive fib calling back and forth between Python & Julia. Only problem, Python packages still a nightmare to install (we're working on that too).
    3 korisnika označavaju da im se sviđa
    Poništi
  6. 22. sij

    Have you ever wondered what will be the ML frameworks of the '20s? In this essay, I examine the directions AI research might take and the requirements they impose, concluding with an overview of what I believe to be two strong candidates: JAX and S4TF.

    13 replies 128 proslijeđenih tweetova 521 korisnik označava da mu se sviđa
    Poništi
  7. proslijedio/la je Tweet
    16. sij
    Odgovor korisniku/ci

    Spaced repetition shines as a tool for memorization, not learning. I have over 10k cards in Anki that range from useful VIM motions I rarely use to common tricks for solving differential equations. I can't imagine trying to maintain all of this information without it.

    1 proslijeđeni tweet 3 korisnika označavaju da im se sviđa
    Poništi
  8. 16. sij

    I'm a 30 years old adult(-ish), but this cute little game has been by far the most efficient way to learn and review VIM's quirks and features.

    1 proslijeđeni tweet 4 korisnika označavaju da im se sviđa
    Poništi
  9. 13. sij

    Did you know that 2.1 was released this weekend? This seems like a great time to "remaster" my DRL with TF2 blog post, so... here it is! I basically rewrote the whole background section and tweaked the code in many ways (fixed a mem leak :o)

    1 reply 7 proslijeđenih tweetova 24 korisnika označavaju da im se sviđa
    Poništi
  10. 23. pro 2019.

    When I decided to switch into AI, I quickly realized I needed strong math foundation via higher education (I was self-taught at the time). Surprisingly many people I talked to thought it was a colossal waste of time, but now 6 years later I'm very happy I didn't listen.

    👇The myths about high-school dropouts are incorrigible. Although one or two genius break through here and there I seriously doubt this is the way to go. I suspect for most people a path to success is through (often lifelong) climbing one step after another.
    Prikaži ovu nit
    1 proslijeđeni tweet 16 korisnika označava da im se sviđa
    Poništi
  11. 23. pro 2019.

    Consistently studying/working is great for expanding existing knowledge, but taking time off to just try out random things has always brought me great returns in the long run. In fact most of my career jumps can be traced back to some 1-2 weeks of messing around with a subject.

    1 reply 16 korisnika označava da im se sviđa
    Poništi
  12. 18. pro 2019.

    Just discovered that 's Five and have a similar number of parameters - around 150 million. I wonder if it's just a fun coincidence or there's some deeper connection.

    1 reply
    Poništi
  13. 18. pro 2019.

    PSA: If you're using 2.x + 3.8 and get a cryptic exception about findCaller, update absl-py package to 0.9.0. More info here:

    4 proslijeđena tweeta 5 korisnika označava da im se sviđa
    Poništi
  14. 13. pro 2019.

    Wow, that's quite a detailed paper! The "Surgery" setup looks especially interesting, I remember being baffled by continually training the policy for ~10 months. Also, kudos on the reporting style! Wonder if recent discussions led by contributed. :)

    0:11
    We're releasing "Dota 2 with Large Scale Deep Reinforcement Learning", a scientific paper analyzing our findings from our 3-year Dota project: One highlight — we trained a new agent, Rerun, which has a 98% win rate vs the version that beat .
    5 korisnika označava da im se sviđa
    Poništi
  15. 26. stu 2019.

    Feynman had incredible insight into the limitations of AI! In this 1985 QA he foresaw many if the issues that plague the field today:

    1 proslijeđeni tweet 8 korisnika označava da im se sviđa
    Poništi
  16. 25. stu 2019.

    This is by far the most comprehensive tutorial (more like a book) on internals I've ever seen by

    16 proslijeđenih tweetova 58 korisnika označava da im se sviđa
    Poništi
  17. 21. stu 2019.

    A commendable effort but at the same time kind of insane the lengths Typescript core devs have to go even with seemingly basic features...

    That's what working with the process looks like - we waited over 5 years (an eternity in JS tooling years) before adding our most-wanted feature because we're *not* taking over an open standard.
    Prikaži ovu nit
    Poništi
  18. 21. stu 2019.

    And it's a wrap! Was quite fun & exciting to be sharing knowledge and some career guidance with the newcomers of If anyone's curious, slides in English are available at

    1 proslijeđeni tweet 5 korisnika označava da im se sviđa
    Poništi
  19. 20. stu 2019.

    Tommorow I'll have the honor of giving a talk on AI & DRL to first-year students of . I think I'm more nervous than when talking to a room full of professionals. Hopefully I'll be able to inspire our young minds to become the AI researchers of the future!

    1 reply 9 korisnika označava da im se sviđa
    Poništi
  20. 7. stu 2019.

    Nice to see that Serral has validated MMR estimates! I think placing AS ~500 MMR below him is fair, people have read too much into a couple of low stakes for fun games.

    Odgovor korisniku/ci @jaakkosarela
    It was okay, I doubt i would lose too many games with a proper setup. I think the 6.3-6.4 mmr is pretty accurate, so not bad at all but nothing special at the same time.
    Poništi

@Inoryy još nije objavio nijedan Tweet.

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:

    ·