Tom Le Paine

@TomLePaine

Research Scientist at DeepMind. PhD from UIUC. Playing with deep learning. Tweets reflect my views alone.

Vrijeme pridruživanja: listopad 2015.

Tweetovi

Blokirali ste korisnika/cu @TomLePaine

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

  1. Prikvačeni tweet
    30. lis 2019.

    Excited to share our work. It's incredible how far demonstrations and reinforcement learning can go!

    Poništi
  2. proslijedio/la je Tweet
    29. sij

    Meena was simply a name that dreamt of, but if you'd like an acronym, here's mine: "May End-to-End Neural-network be AGI"! thanks to all great collaborators, David, , , , Zi, , , , and !

    Poništi
  3. proslijedio/la je Tweet
    16. pro 2019.

    Are you working at the intersection of cogsci, neuro and AI? Want a chance to meet and discuss with others working at this intersection? Then submit something to our workshop on "Bridging AI and Cognitive Science" at ! Deadline is Feb. 5th

    Poništi
  4. proslijedio/la je Tweet
    26. pro 2019.

    So a big implementation challenge for DL is to figure out how to get symbol processing within a differentiable programming/continuous activation implementation. 3/

    Prikaži ovu nit
    Poništi
  5. proslijedio/la je Tweet

    So incredible to see and the amazing game of featured on the cover of ! GG!

    Poništi
  6. proslijedio/la je Tweet
    6. stu 2019.

    Naturally, this is a single-authored paper: "Fast Transformer Decoding: One Write-Head is All You Need," Noam Shazeer:

    Poništi
  7. proslijedio/la je Tweet

    : Grandmaster level as all 3 races on , w/ a pro approved interface (camera & APM limits). 2 years ago I thought this was impossible! How? Imitation learning (Diamond) -> multiagent League (Grandmaster)

    Prikaži ovu nit
    Poništi
  8. proslijedio/la je Tweet
    16. lis 2019.

    We have shown that it is possible to train very large transformers on RL problems by introducing gating mechanism to stabilize them. It turned out this architecture is very effective on a number of RL benchmarks. As pointed out,mostly due to the work done by Emilio.

    Poništi
  9. proslijedio/la je Tweet
    27. ruj 2019.

    How can robots learn from humans and their own experience to manipulate objects using vision? Here is our take on the problem:

    Prikaži ovu nit
    Poništi
  10. 24. ruj 2019.

    When you finish a PhD in AI, they put you in a special room & tell you everything you've experienced is part of a simulation. But even worse, you aren't the proposed method, just a baseline.

    Poništi
  11. proslijedio/la je Tweet
    5. ruj 2019.

    R2D3 uses demonstrations to solve hard exploration problems in partially observable environments with highly variable initial conditions. It can solve tasks where SOTA methods fail to see a single reward. Paper: Videos & more:

    Poništi
  12. proslijedio/la je Tweet
    5. ruj 2019.

    Advancing the frontier of what deep RL and imitation methods can achieve. Congratulations and the many other brilliant researchers and engineers at who worked on this ambitious project for many months.

    Poništi
  13. proslijedio/la je Tweet

    StarCraft update! As part of ongoing scientific research into artificial intelligence, players can now opt-in for the chance to be matched against experimental versions of DeepMind's StarCraft II agent, AlphaStar, on the competitive ladder. Learn more:

    Poništi
  14. proslijedio/la je Tweet
    25. lip 2019.
    Prikaži ovu nit
    Poništi
  15. proslijedio/la je Tweet
    11. lip 2019.

    Incredibly honoured to have received an honourable mention for best paper at ICML this year 😊

    Prikaži ovu nit
    Poništi
  16. proslijedio/la je Tweet
    5. lip 2019.

    Deep Learning's greatest virtue is that the community is relentlessly empirical. It doesn't matter if your method is "bad" or "messy" or "wrong" or "unprincipled." If you can solve a problem that people care about, then they will listen to you.

    Poništi
  17. proslijedio/la je Tweet
    8. tra 2019.

    Excited to share our newest work (& my first final-author paper!) led by Victor Bapst and Alvaro Sanchez-Gonzalez (w/ ). We investigate structured, object-centric agents that can build things!👷‍♀️🏗

    Prikaži ovu nit
    Poništi
  18. proslijedio/la je Tweet
    4. ožu 2019.

    Our work on model-based RL for Atari using video prediction. While SimPLe in theory, there are a lot of details that matter! Luckily, easy to reproduce with the code that is available at . Also make sure to check out the videos at

    Poništi
  19. proslijedio/la je Tweet
    4. ožu 2019.

    Goodfellow sits down and ponders How to beat autoencoders? Oh, what a man! He invented GAN To generate nice bedroom pictures

    Poništi
  20. 21. velj 2019.

    Closest thing I've had to a celebrity doppelgänger: half white, half South East Asian, with a master's in engineering and a Japanese barber

    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:

    ·