Bo Dai

@daibond_alpha

Research Scientist at Google Brain

California, USA
sites.google.com/site/daibohr/
Vrijeme pridruživanja: listopad 2012.
1 fotografija ili videozapis Fotografije i videozapisi

Tweetovi

Blokirali ste korisnika/cu @daibond_alpha

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

  1. 24. sij

    I actually dug into this paper after reading a blog from Ben Recht about almost the same thing.

    This paper from 1988 by LeCun is probably one of the first to introduce Pontryagin's maximum principle to ML. Notably, the adjoint sensitivity method used to train neural ODEs is also based on this principle. The adjoint (co-state) variable is basically a Lagrange multiplier.
    3 proslijeđena tweeta 14 korisnika označava da im se sviđa
    Poništi
  2. proslijedio/la je Tweet
    13. sij

    "To honour the memory of those lost, the University of Toronto has established an Iranian Student Memorial Scholarship Fund." The first $250K donations is matched 3:1 by . Any funds received beyond $250K will be matched at a rate of 1:1.

    64 proslijeđena tweeta 215 korisnika označava da im se sviđa
    Poništi
  3. proslijedio/la je Tweet
    10. sij

    A tour de force by Abbe & Sandon, "Any function distribution that can be learned from samples in poly-time can also be learned by a poly-size neural net trained with SGD on a poly-time initialization with poly-steps" + "[this] does not hold for GD"

    1 reply 23 proslijeđena tweeta 110 korisnika označava da im se sviđa
    Poništi
  4. proslijedio/la je Tweet
    8. sij

    Many of our group's recent papers - DualDICE, ValueDICE, GenDICE, AlgaeDICE - can be framed as applications of this duality. Still, lots of potential remaining applications for others to explore, and lingering questions of how these formulations interplay with stoch. opt methods.

    Fenchel-Rockafellar duality is a powerful tool that more people should be aware of, especially for RL! Straightforward applications of it enable offpolicy evaluation, offpolicy policy gradient/imitation learning, among others
    4 proslijeđena tweeta 14 korisnika označava da im se sviđa
    Poništi
  5. 8. sij

    We unified the DICE family via a primal-dual perspective, from which we can generate more variants of DICE for different RL problems!

    Fenchel-Rockafellar duality is a powerful tool that more people should be aware of, especially for RL! Straightforward applications of it enable offpolicy evaluation, offpolicy policy gradient/imitation learning, among others
    25 korisnika označava da im se sviđa
    Poništi
  6. 13. pro 2019.

    Welcome to our workshop tomorrow. See you in west ballroom A!

    We just announced the best paper awards of the OptRL workshop: . Come discover these great works and others tomorrow in West ballroom A!
    Prikaži ovu nit
    2 proslijeđena tweeta 10 korisnika označava da im se sviđa
    Poništi
  7. 8. pro 2019.

    Congratulations to Lin!

    Congratulations to our colleague Lin Xiao for the test of time award!!! Online convex optimization and mirror descent for the win!! (As always? :-).)
    2 korisnika označavaju da im se sviđa
    Poništi
  8. 4. pro 2019.

    The schedule and accepted papers are released: . Congratulations to all the recipients of the travel awards. We thank all the invited speakers, panelists and authors. Thanks to our sponsors and . See you in Vancouver next week.

    Our workshop proposal (with , , Bo Dai, Niao He and Dale Schuurmans) on the optimization foundations of RL has been accepted to NeurIPS. I can't wait to see what the community is doing at the intersection of these two fields.
    6 proslijeđenih tweetova 40 korisnika označava da im se sviđa
    Poništi
  9. 4. pro 2019.

    We reveal the secret of the “DualDICE trick” via an elegant primal-dual lens, which generalize the trick to more general application areas :)

    Very excited about my new paper! We formulate the on-policy max-return RL objective w.r.t *arbitrary* offline data and without *any* explicit importance correction. Amazingly, the gradient of the objective w.r.t pi is exactly the on-policy policy gradient!
    7 proslijeđenih tweetova 18 korisnika označava da im se sviđa
    Poništi
  10. 3. pro 2019.

    Here is our poster: We also released the code with some animation:

    2 proslijeđena tweeta 23 korisnika označavaju da im se sviđa
    Poništi
  11. proslijedio/la je Tweet
    20. stu 2019.

    "Mastering Atari, Go, Chess and Shogi by Planning with a Learned Model" -- Really exciting new work from the AlphaZero group at DeepMind!

    1 reply 29 proslijeđenih tweetova 91 korisnik označava da mu se sviđa
    Poništi
  12. proslijedio/la je Tweet
    5. svi 2019.

    Q: "Should I learn a density model? Or should I learn how to draw samples?" A: "Yes"

    34 proslijeđena tweeta 195 korisnika označava da im se sviđa
    Poništi
  13. proslijedio/la je Tweet
    4. stu 2019.

    rely on approximate sampling algorithms, leading to a mismatch between the model and inference. Instead, we consider the sampler-induced distribution as the model of interest yielding a class of tractable . ()

    24 proslijeđena tweeta 129 korisnika označava da im se sviđa
    Poništi
  14. proslijedio/la je Tweet
    23. lis 2019.

    Very nice blog post from Song Mei on the replica method from statistical physics.

    71 proslijeđeni tweet 284 korisnika označavaju da im se sviđa
    Poništi
  15. proslijedio/la je Tweet
    25. ruj 2019.

    This new book: Handbook of Graphical Models captures well how the field has grown since its inception in the 1980's. Written by top statisticians, it can serve as a good introduction to causal modeling for traditionally-trained statisticians.

    159 proslijeđenih tweetova 701 korisnik označava da mu se sviđa
    Poništi
  16. proslijedio/la je Tweet
    23. srp 2019.

    Our workshop proposal (with , , Bo Dai, Niao He and Dale Schuurmans) on the optimization foundations of RL has been accepted to NeurIPS. I can't wait to see what the community is doing at the intersection of these two fields.

    15 proslijeđenih tweetova 127 korisnika označava da im se sviđa
    Poništi

@daibond_alpha 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:

    ·