Tweetovi

Blokirali ste korisnika/cu @anirudhg9119

Jeste li sigurni da želite vidjeti te tweetove? Time nećete deblokirati korisnika/cu @anirudhg9119

  1. Prikvačeni tweet
    25. ruj 2019.

    This is what I've been doing for quite some time. Recurrent Independent Mechanisms (RIMs) , joint work with Alex Lamb, , , , and Yoshua Bengio. Here's a short summary:

    Prikaži ovu nit
    Poništi
  2. 14. sij

    Invited speakers include Yoshua Bengio, , , Lars Buesing, , , Christina Heinze-Deml.

    Poništi
  3. 9. sij

    We are organizing a workshop on Causal learning for Decision Making at along with , Jovana Mitrovic, , Stefan and . Consider submitting your work!

    Poništi
  4. 4. sij

    Thanks! Previous work like Outrageously Large Neural Networks () as well as Sparse Attentive Backtracking () introduced have also used dynamic sparse attention.

    Poništi
  5. 23. pro 2019.

    This paper got accepted at ICLR'20. Would not have been possible without , Jonathan Binas, , and Yoshua Bengio. . More feedback is very much appreciated :)

    Poništi
  6. 22. pro 2019.

    See, also the followup to this led by : )

    Prikaži ovu nit
    Poništi
  7. 22. pro 2019.

    This work also got accepted at . "A meta-transfer objective for learning to disentangle causal mechanisms" Led by Yoshua Bengio.

    Prikaži ovu nit
    Poništi
  8. 22. pro 2019.

    This paper got accepted at . Great work by . Along with Steffen Wolf, Roman Remme, Yoshua Bengio. "Learning the Arrow of Time for Problems in Reinforcement Learning"

    Poništi
  9. 21. pro 2019.

    Two things which I really enjoy : 1. Going through reviews of accepted papers. 2. Going through papers which got "rejected", and understanding what other groups are doing ;)

    Poništi
  10. 20. pro 2019.

    Lesson learned: If you have reviewers, who are not confident, then during the review process explicitly ask AC to assign a new reviewer. One of the papers got rejected with 3 weak accepts, and in discussion reviewers said they are satisfied but PCs overruled.

    Poništi
  11. 14. pro 2019.

    Another reminder to read, Parallel Distributed Processing (PDP) book, :)

    Prikaži ovu nit
    Poništi
  12. 14. pro 2019.

    "Extending Machine Language Models toward Human-Level Language Understanding" by Jay McClelland, with , , Maja Rudolph, Hinrich Schutze. is interesting read (Even if you aren't interested in language)

    Prikaži ovu nit
    Poništi
  13. 13. pro 2019.

    Environments must allow sufficient causal systematically to allow transfer of information.

    Prikaži ovu nit
    Poništi
  14. 13. pro 2019.

    I strongly agree with this. Data distribution should be growing (adding more tasks more time) , composite (combination of two or more sub-tasks) as well as diverse (various different tasks).

    Prikaži ovu nit
    Poništi
  15. 7. pro 2019.

    Yoshua Bengio talking about "why backprop is not biologically plausible"

    Poništi
  16. 3. pro 2019.

    1) Infobot: 2) Sparse Attentive Backtracking 3) Recurrent Independent Mechanisms

    Prikaži ovu nit
    Poništi
  17. 3. pro 2019.

    Lot of our work has been in combining system 1 and system 2, but system 2 is also implemented within the framework of gradient based learning, and incorporating such ideas seems to improve generalization.

    Prikaži ovu nit
    Poništi
  18. 29. stu 2019.

    Well, many researchers understand this and are actively trying to build systems which has these inductive biases.

    Poništi
  19. 29. stu 2019.

    Once this happens, the life-force sustaining the debate vanishes, and only shitstorms are left. (2/2)

    Prikaži ovu nit
    Poništi
  20. 29. stu 2019.

    Civilized debates start off with certain common axioms shared by all participants. When they are not, and people rely on different or even opposite axioms, people end up talking right past each other, and soon someone will believe that the other has bad faith. (1/2)

    Prikaži ovu nit
    Poništi
  21. 26. stu 2019.

    This is probably the most succinct summary of various ways in which causality could be useful for machine learning by Highly recommended.

    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:

    ·