Ida Momennejad

@criticalneuro

Cognitive Computational Neuroscience , prev. . Reinforcement learning, memory, planning, collective memory. fMRI, electrophys. Makes music 🎹

New York, NY
Vrijeme pridruživanja: prosinac 2014.

Medijski sadržaj

  1. 3. velj
    Odgovor korisniku/ci

    2017:Successor Features for Transfer in RL 2019:Universal Successor Features for Transfer Reinforcement Learning 2020:Fast Task Inference with Variational Intrinsic Successor Features Robotics: Visual Semantic Planning using Deep Successor Representations

  2. 4. sij

    Hi friends. Inspired by Tolman, who was vocally political, here’s a political tweet against a war with Iran. We held a peaceful protest near prospect park in Brooklyn where senator lives. We hope he reflects the anti war voices of NY. Next: Tomorrow 11am time square.

  3. 27. pro 2019.
    Odgovor korisniku/ci

    Representing cog comp neuro here. Note that there's also critiques/revisions to the three levels in cognitive science. A notable critique comes from Danks, arguing that cognitive theory is not adequately captured by Marr's levels, arguing for "constraint" rather than reduction.

  4. 20. pro 2019.
    Odgovor korisniku/ci
  5. 19. pro 2019.

    The videos of our talks at the workshop on "Biological and Artificial Reinforcement Learning" are online. Many thanks to :) If you were so inclined, you can find my talk at 1:02:15 in this link:

  6. 13. pro 2019.

    Superdyna is here! An integrated approach to AI agents who can find their way around the world by discovery and aberration. :)

    Prikaži ovu nit
  7. 13. pro 2019.

    Inspired by play, we need a set of good subproblems. They come from state features. State feature attainment is a distinct subproblem. 5/n

    Prikaži ovu nit
  8. 13. pro 2019.

    Expectation planning better than other approaches to looksheads and updateds. 4/n

    Prikaži ovu nit
  9. 13. pro 2019.

    This is a follow up to the 1990 Dyna that works like this: 3/n

    Prikaži ovu nit
  10. 13. pro 2019.

    The world is complex and super dyna needs to handle partial observations and online TD learning to extract options and build a world model. 2/n

    Prikaži ovu nit
  11. 13. pro 2019.

    Will tweet about Rich Sutton’s talk on SuperDyna :) @

    Prikaži ovu nit
  12. 11. pro 2019.

    Excellent talk by Eszter Vertez (with Maneesh Sahani) on a neurally plausible model learning successor representation in partially observable environments.

  13. 11. pro 2019.
    Odgovor korisniku/ci
  14. 11. pro 2019.

    Not the robot we wanted, but the robot we deserved. 🤖@

  15. 8. pro 2019.

    Finally met in person 💜 Need more of this=> champagne 🍾 food & lots of dessert 🍨 with past & future coauthors Mina, ! So much fun! It fueled my 7 total hours of talk & python workshop over 2 days &💥brainstorms! Can’t wait for next phase!

  16. 7. pro 2019.
    Odgovor korisniku/ci
  17. 3. pro 2019.
    Odgovor korisnicima

    Same! For me it’s like a split personality: OCD about typos when proofreading someone else’s work and dyslexic when it comes to proofreading my own!

  18. 2. pro 2019.
    Odgovor korisniku/ci
  19. 2. pro 2019.

    After extensive experience with university admin in four different countries, I am profoundly grateful for the special level of patience, kindness, & support I have received from admins in US universities both at Princeton & Columbia. Without them I would have drowned in paper!🙏🏼

  20. 30. stu 2019.
    Odgovor korisnicima i sljedećem broju korisnika:

    Lol confounded design -> over-the-wall & under-the-stairs goal vectors

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:

    ·