Tweetovi

Blokirali ste korisnika/cu @ReedRoof

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

  1. proslijedio/la je Tweet
    1. velj

    Rotation, Translation, and Cropping for Zero-Shot Generalization Makes a lot of sense. Try playing Doom not from an agent-centric perspective! I think agent-centric view is a better prior for encoding useful information using fewer bits for the policy.

    Poništi
  2. proslijedio/la je Tweet
    26. sij

    Quaternions and Euler angles are discontinuous and difficult for neural networks to learn. They show 3D rotations have continuous representations in 5D and 6D, which are more suitable for learning. i.e. regress two vectors and apply Graham-Schmidt (GS).

    Prikaži ovu nit
    Poništi
  3. proslijedio/la je Tweet
    23. sij

    We released the code and data for GraspNet paper (). Code and data can be found at .

    Poništi
  4. proslijedio/la je Tweet
    27. sij

    TRADI: Tracking deep neural network weight distributions -- work with G. Franchi We’re proposing a cheap method for getting ensembles of networks from a single network training 1/

    Prikaži ovu nit
    Poništi
  5. proslijedio/la je Tweet
    26. sij

    Teaching Deep Unsupervised Learning (2nd edition) at this semester. You can follow along here: Instructor Team: , , , Wilson Yan, Alex Li, YouTube, PDF, and Google Slides for ease of re-use

    Poništi
  6. proslijedio/la je Tweet
    21. sij

    Facebook AI has effectively solved the task of point-goal navigation by AI agents in simulated environments, using only a camera, GPS, and compass data. Agents achieve 99.9% success in a variety of virtual settings, such as houses and offices.

    Poništi
  7. proslijedio/la je Tweet

    Every 1-2 months, "QT-Opt grasping" paper authors receive an email asking about implementation details of the convnet. Took awhile, but we finally got around to open-sourcing the QT-Opt model as a Tensor2Robot model.

    Prikaži ovu nit
    Poništi
  8. proslijedio/la je Tweet
    23. sij

    Q-learning is difficult to apply when the number of available actions is large. We show that a simple extension based on amortized stochastic search allows Q-learning to scale to high-dimensional discrete, continuous or hybrid action spaces:

    Poništi
  9. proslijedio/la je Tweet
    23. sij

    Great coverage by in regarding our recent work on effective treatment of hybrid problems with discrete-continuous action spaces in their native form (imagine e.g. controlling gears and gas in your car).

    Poništi
  10. proslijedio/la je Tweet
    23. sij

    Today, we open-sourced our Spot software development kit and announced our first-ever user conference Actuate! Learn more about what users like and are doing with Spot's SDK on our blog: See the full video:

    Poništi
  11. proslijedio/la je Tweet
    23. sij

    We have two papers on learning keypoint representations for robot manipulation accepted in . 6-PACK: Category-level 6D Pose Tracker with Anchor-Based Keypoints KETO: Learning Keypoint Representations for Tool Manipulation

    Poništi
  12. proslijedio/la je Tweet
    22. sij

    Excited to share PCGrad, a super simple & effective method for multi-task learning & multi-task RL: project conflicting gradients On Meta-World MT50, PCGrad can solve *2x* more tasks than prior methods w/ Tianhe Yu, S Kumar, Gupta, ,

    Poništi
  13. proslijedio/la je Tweet
    18. sij

    Check out this awesome project that BAIR student Greg Kahn () worked on at on training an autonomous deep neural network pilot to film while avoiding obstacles!

    Poništi
  14. proslijedio/la je Tweet
    1. sij

    Can suboptimal trials serve as optimal demos? Suboptimal trials are "optimal" for a policy that *aims* to be suboptimal. By conditioning policy on the reward (or advantage) we want it to get, we can use all trials as demos: w/ Aviral Kumar &

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

    Can robots learn about the world by observing humans? Learn to predict with both interaction & observation (of humans), then use the model to accomplish goals.
 w. Schmeckpeper , Xie, , Tian, ,

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

    It’s time to get rid of / tpu / cuda dependencies. On premises training with a new end-to-end framework is a solution if the dev put efforts on it. While is production ready and is a choice for they both miss flexibility.

    Poništi
  17. proslijedio/la je Tweet
    25. pro 2019.

    Video & slides for LIRE workshop @ are now up: Check out the Talks and Panel by Jeff Bilmes Tom Griffiths & more. Thanks to all speakers & presenters for making the workshop a success!

    , , i još njih 3
    Poništi
  18. proslijedio/la je Tweet
    18. pro 2019.

    A new tasked with opening pill bottles, designed by scientists, could also explain its actions in multiple ways while it performed the task. Learn more about this study, funded by 's explainable AI program:

    Poništi
  19. proslijedio/la je Tweet
    19. pro 2019.

    Can we discover structure & meta-learn across it in unsegmented time series data? MOCA simultaneously detects changepoints & meta-learns across time for continuous adaptation Continuous Meta-Learning without Tasks w , Sharma,

    Poništi
  20. proslijedio/la je Tweet
    18. pro 2019.

    This is what recognition datasets should be (if they have any value at all)... easy for humans, hard for machines!

    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:

    ·