Swaroop Guntupalli

@swaroopgj

Neuroscientist

Hanover, NH & Stanford, CA
Vrijeme pridruživanja: veljača 2014.

Tweetovi

Blokirali ste korisnika/cu @swaroopgj

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

  1. proslijedio/la je Tweet
    6. pro 2019.

    You'd think a generative model should answer different queries but deep learning models like VAEs are trained to answer just one type of query! We introduce a method called query-training that creates an inference network that can answer novel query types.

    Poništi
  2. proslijedio/la je Tweet
    7. pro 2019.

    How is it even possible, now that "AI has taken over", that I buy a washing machine and everything starts recommending me more washing machines? (don't get me wrong: it is a comforting feeling that, outside the movies, not much has happened since Eliza)

    Poništi
  3. proslijedio/la je Tweet
    6. pro 2019.

    We show that a singular principle of variable-order sequence learning in a structured model can explain cognitive maps phenom: spatial maps, transitive inference, temporal order, hierarchical planning, remapping etc. -- all important for AI & neuroscience.

    Poništi
  4. proslijedio/la je Tweet
    5. pro 2019.

    Great interview, in which the man himself explains how he manages to get an infinite amount of work done in the same finite time that the rest of us have ;-) thx !

    Poništi
  5. proslijedio/la je Tweet
    13. ruj 2019.

    It also learns a lossy compressed graph representation of the underlying sequences. Closely related to our 1st paper on cognitive maps as higher-order sequences.

    Prikaži ovu nit
    Poništi
  6. 5. ruj 2019.
    Poništi
  7. proslijedio/la je Tweet
    27. kol 2019.

    Our paper proposes a new model for cognitive maps in the hippocampus which involves learning higher-order graphs of event sequences. It can explain flexible planning, route-dependence of place cells, and subsumes successor representations.

    Poništi
  8. proslijedio/la je Tweet
    18. srp 2019.

    1/ New in w/ lab: : new opsin + multi-photon holography to image ~4000 cells in 3D volumes over 5 cortical layers while also stimulating ~50 neurons to directly drive visual percepts; data analysis and theory reveal…

    Prikaži ovu nit
    Poništi
  9. proslijedio/la je Tweet
    7. svi 2019.

    Thread about our new paper on learning the structure of variable order sequences. Imposing a biologically-inspired sparsity structure alleviates the credit diffusion problem in HMMs! Also relevant for neuroscience... (1)

    Prikaži ovu nit
    Poništi
  10. proslijedio/la je Tweet
    6. svi 2019.

    If yuor laugnage moedl cnnaot raed tihs, it mgiht not be vrey dnagorues 😜. See our paper on learning graphs that can model higher-order sequences and deal with uncertainty.

    Poništi
  11. proslijedio/la je Tweet
    22. sij 2019.

    A thought is a program! Check out our blog to see how we are building a cognitive architecture towards AGI. Also included: code, data, and a free link to our recent Science Robotics paper on concept learning.

    Poništi
  12. proslijedio/la je Tweet
    17. sij 2019.

    Mini thread about the cognitive science and neuroscience inspirations behind our new paper in which we learn concepts as 'cognitive programs' on a 'visual cognitive computer'.

    Prikaži ovu nit
    Poništi
  13. proslijedio/la je Tweet

    Instead of relying on a list or rules, a new computational framework for learning lets robots come up with their own concepts by detecting abstract differences in images and then recreating them in real life:

    Poništi
  14. proslijedio/la je Tweet
    16. sij 2019.

    Our concept-learning work is now published in Science Robotics! Checkout this very nice video that Science made to explain the paper.

    Poništi
  15. proslijedio/la je Tweet
    10. pro 2018.

    Our new paper learns concepts as programs on a 'visual cognitive computer', showing zero-shot task-transfer on robots, and strong generalization. We bring cogsci ideas about perceptual symbol systems and image schemas into the realm of machine learning.

    Poništi
  16. proslijedio/la je Tweet

    What a gift! Ramanujan’s papers, now online.

    Poništi
  17. proslijedio/la je Tweet
    11. ruj 2018.

    Moore's law sputtering to an end. Now is the time for new architectures and creativity!

    Poništi
  18. proslijedio/la je Tweet
    11. ruj 2018.

    New paper with , , and ! Hyperalignment factors out idiosyncrasies in functional–anatomical correspondence, revealing individual differences in fine-grained functional architecture otherwise obscured by anatomical normalization

    Poništi
  19. proslijedio/la je Tweet
    7. kol 2018.

    Checkout our CCN-18 papers! (1) RCN-derived circuits suggest precise functional roles for feed-forward, feedback, and lateral connections, and for the thalamus. (2) RCN can explain multiple visual phenomena including two well-known illusions.

    Prikaži ovu nit
    Poništi
  20. proslijedio/la je Tweet
    7. kol 2018.

    Here are the links to our CNN-18 papers: Cortical Microcircuits from a Generative Vision Model: Explaining Visual Cortex Phenomena using Recursive Cortical Network:

    Prikaži ovu nit
    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:

    ·