Julian Togelius

@togelius

AI and games researcher. Associate professor at NYU; Editor-in-Chief of ; director of ; co-founder of .

New York City
Vrijeme pridruživanja: siječanj 2009.
Rođen/a 1979.

Tweetovi

Blokirali ste korisnika/cu @togelius

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

  1. Prikvačeni tweet
    28. sij

    What if we see level generation as a game, where each action changes the level in some way? Well, we could use RL to learn to "play" level generation! We introduce Procedural Content Generation via Reinforcement Learning (PCGRL), a new paradigm for PCG.

    Prikaži ovu nit
    Poništi
  2. prije 16 sati

    This is the kind of artist I would be if I was an artist.

    Poništi
  3. proslijedio/la je Tweet
    prije 23 sata

    Wow: Google's "Meena" chatbot was trained on a full TPUv3 pod (2048 TPU cores) for **30 full days** - That's more than $1,400,000 of compute time to train this chatbot model. (! 100+ petaflops of sustained compute !)

    Poništi
  4. proslijedio/la je Tweet
    prije 22 sata

    Dear fellow conference paper reviewers: Smith (2009) without including the full reference is not helpful feedback. Take a second copy and paste the reference from Google Scholar if you really want to help the author, rather than make you feel smug about yourself.

    Poništi
  5. proslijedio/la je Tweet

    it's faculty job hunting season! i wrote up the advice i give my students and others as they start interviewing. tips are mostly relevant to HCI/CS interviews ::

    Poništi
  6. proslijedio/la je Tweet
    1. velj

    To the left, you see a trained agent playing a level of a game. To the right, you see the same playthrough from an agent-centric perspective: cropped, translated, and rotated with the agent in the center. Which perspective is the best input for the agent?

    Prikaži ovu nit
    Poništi
  7. 1. velj

    In any case, even if this is possible in principle, it seems that the way we represent the input makes a lot of difference for the generality of skills that can be learned in practice.

    Prikaži ovu nit
    Poništi
  8. 1. velj

    It is possible that the standard paradigm of a neural network with a handful of layers learning to master, say, Atari games from a static third-person view is actually impossible. That is, it doesn't learn any general playing skills. It learns some kind of stimulus-response table

    Prikaži ovu nit
    Poništi
  9. 1. velj

    Also, can it? It is hard to imagine that a neural network of just a few layers could actually implement the transformations necessary to even understand where things are relative to the agent, so that the policy can be location-independent?

    Prikaži ovu nit
    Poništi
  10. 1. velj

    The promise/premise of deep learning is that we don't have to worry about these representations, because the network will figure out the input representation itself. But, really, will it? Will a network of a few layers really learn to rotate and translate to focus on the agent?

    Prikaži ovu nit
    Poništi
  11. 1. velj

    It is well known from "non-deep" agent learning research (say, evolutionary robotics work and game AI work from early to mid 2000s) that the sensor representation is extremely important to the agent's ability to learn.

    Prikaži ovu nit
    Poništi
  12. 1. velj

    This paper builds on earlier work, where we showed that standard deep RL algorithms learn policies that generalize very badly. They are barely able to play any levels that they were not trained on at all.

    Prikaži ovu nit
    Poništi
  13. 1. velj

    Note that in all six conditions, the algorithm is able to learn policies that work on the specific level(s) it is trained on. What really differs is the generalization capacity.

    Prikaži ovu nit
    Poništi
  14. 1. velj

    Simply cropping the level to see only what's around the agent, or rotating it so that the agent is pointing up, or translating it so that the agent is always in the center of the image, have little effect on their own. But all three together drastically increases generalization!

    Prikaži ovu nit
    Poništi
  15. 1. velj

    We use a standard deep network architecture and reinforcement learning algorithm (A2C). When agents are trained using the top-down view which is the game's "native" view (to the left in the inital gif) the trained networks play unseen levels very badly.

    Prikaži ovu nit
    Poništi
  16. 1. velj

    In our new paper, "Rotation, Translation, and Cropping for Zero-Shot Generalization" by and I show that the agent-centric perspective is better in the sense that the agent learns policies that generalize better.

    Prikaži ovu nit
    Poništi
  17. 1. velj

    To the left, you see a trained agent playing a level of a game. To the right, you see the same playthrough from an agent-centric perspective: cropped, translated, and rotated with the agent in the center. Which perspective is the best input for the agent?

    Prikaži ovu nit
    Poništi
  18. 30. sij

    "It is a truth universally acknowledged, that a single agent in possession of a good reward, must be in want of an environment." (This sentence was just deleted from the ICML abstract that and I are writing)

    Poništi
  19. proslijedio/la je Tweet
    30. sij

    Legitimately one of the best insights I've internalized after years of doing research is the old classic - "ideas are cheap, execution is everything" [not quite everything, we still need cool ideas, but y'know]

    Poništi
  20. proslijedio/la je Tweet
    30. sij

    Amazing: every week I see a paper comparing algorithms using mean performance over *3* seeds ! Yes, ****3**** !!! Please please community, your great ideas will be served better using standard scientific methods!

    Poništi
  21. proslijedio/la je Tweet
    30. sij

    imagine how much more comprehensible academic papers would be if literacy in visual communication was as universal as literacy in writing idk why we treat diagrams & illustrations as subordinate to the written parts of papers when they’re often more important for comprehension

    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:

    ·