Torsten Scholak

@tscholak

applied research scientist , formerly physicist , into natural language processing, Bayesian stats, and functional programming in

Montréal
Vrijeme pridruživanja: veljača 2010.

Tweetovi

Blokirali ste korisnika/cu @tscholak

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

  1. Prikvačeni tweet
    23. pro 2019.
    Odgovor korisniku/ci

    Here’s a fully functional convolutional neural network in Haskell. This API wraps libtorch, and the compiler checks your shapes, data types and physical devices, and it tells you how functions change them (see the hover box for unsqueeze).

    Poništi
  2. 30. sij

    the n-ary version of this could be achieved with trchniques introduced by in his n-ary-functor library

    Poništi
  3. proslijedio/la je Tweet
    30. sij

    Timely post on Jax’s vmap in Haskell by . Will have to give these ideas a try in !

    Poništi
  4. proslijedio/la je Tweet
    24. sij

    Thanks for the shout out :) It’s true that differentiable programming tech stacks take substantial effort. They also define the space of what gets built and researched. In this way they’re sort of a “narrow waist” in the systems sense.

    Prikaži ovu nit
    Poništi
  5. 23. sij

    I am convinced that the answer to this question is yes.

    Poništi
  6. 23. sij

    I wonder if and how this could be combined with neural approaches for program synthesis, e.g. and 's .

    Poništi
  7. proslijedio/la je Tweet
    20. sij

    Always thought that generic-lens was a shining library usability-wise but a straightforward application of GHC Generics. But after looking at the implementation... It is truly a work of art under the hood, w a LOT more going on than I had thought

    Poništi
  8. 19. sij

    there's a typo, the output type should have been `IO output`

    Prikaži ovu nit
    Poništi
  9. 19. sij

    after a few days of searching for the right abstractions, I got data parallelism working in

    Prikaži ovu nit
    Poništi
  10. 19. sij

    It is absolutely astonishing how quickly was able to put this together: has fully migrated to libtorch 1.4.0! Thank you Junji!

    Poništi
  11. proslijedio/la je Tweet
    18. sij

    I found it very important to learn basics of LISP to start understanding symbolic AI literature. It seems like this programming language for many decades structured the way people thought and communicated with each other.

    Prikaži ovu nit
    Poništi
  12. proslijedio/la je Tweet
    18. sij

    type system applied to AI: - encode tensor dimensions in types (as value-types), express matrix calculus and let compiler check ops validity - encode location of tensors in types (Tensor @ CPU or @ GPU) and let the compiler warn that we can't manipulate resources not colocated.

    Prikaži ovu nit
    Poništi
  13. proslijedio/la je Tweet
    17. sij

    👉 users of lang Y are stupid 👉 creators of lang Y are stupid 👉 your lang X is better than * 👉 only X creates safe&sound code If you believe and act on any of the above you burn the bridges I try to build organizing Stop it or Go Away.

    Prikaži ovu nit
    Poništi
  14. proslijedio/la je Tweet
    16. sij

    Fascinating benchmark study on using lambda calculus HOAS in several languages.

    Poništi
  15. 16. sij
    Poništi
  16. 16. sij
    Poništi
  17. 14. sij
    Poništi
  18. proslijedio/la je Tweet
    14. sij

    I often meet research scientists interested in open-sourcing their code/research and asking for advice. Here is a thread for you. First: why should you open-source models along with your paper? Because science is a virtuous circle of knowledge sharing not a zero-sum competition

    Prikaži ovu nit
    Poništi
  19. 13. sij
    Poništi
  20. 12. sij
    Poništi
  21. 11. sij

    *hint* *hint*

    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:

    ·