mat kelcey

@mat_kelcey

Building scrappy robots on my farm Part time ML @ ThoughtWorks Advisor @ Hyperflow & thingc Mentor @ HoistAI Ex Google Brain (Robotics & NLP), Wavii, AWS... OVG

Victoria, Australia
matpalm.com
Vrijeme pridruživanja: ožujak 2009.
1.183 fotografije ili videozapisa Fotografije i videozapisi

Tweetovi

Blokirali ste korisnika/cu @mat_kelcey

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

  1. Prikvačeni tweet
    13. ruj 2019.

    solving cartpole... by evolving the raw bytes of a 1.4KB tflite microcontroller serialised model

    26 proslijeđenih tweetova 155 korisnika označava da im se sviđa
    Poništi
  2. prije 2 sata

    under-over-around-and-throughfitting

    Odgovor korisniku/ci @mat_kelcey
    This ladies and gents is what under-overfitting looks like
    2 korisnika označavaju da im se sviđa
    Poništi
  3. prije 9 sati

    hold onto your hats.... now we're talking!!!

    8 replies 2 proslijeđena tweeta 111 korisnika označava da im se sviđa
    Poništi
  4. prije 13 sati

    !!! THAT'S NOT HOW HTTP PARAMETERS WORK !!!

    6 korisnika označava da im se sviđa
    Poništi
  5. 3. velj

    home brew sharding. don't judge me.

    4 korisnika označavaju da im se sviđa
    Poništi
  6. 2. velj

    oh, forgot about my most important doc of all!! i also have a "to read" doc where i paste links to arxiv papers in the delusional hope that just acknowledging the existence of a paper ( & occasionally reading the abstract ) will automagically transfer all its info into my brain.

    love "An Opinionated Guide to ML Research" !! i'm a big big fan of journals! the ones i keep are.... 1/5
    Prikaži ovu nit
    5 replies 39 korisnika označava da im se sviđa
    Poništi
  7. 2. velj

    TIL about enumerate start...

    3 proslijeđena tweeta 35 korisnika označava da im se sviđa
    Poništi
  8. 2. velj

    following my three-strikes-rule of "if you answer the same question three times in a short period write it down"; here's a simple example of loss masking for sequences with a custom training loop in tf2.0

    4 proslijeđena tweeta 25 korisnika označava da im se sviđa
    Poništi
  9. 1. velj

    Got it!!

    1 proslijeđeni tweet 5 korisnika označava da im se sviđa
    Poništi
  10. 1. velj

    wow. that's some +1000 level 🔥🔥🔥 right there

    Here it is- I Flew Katie Hopkins to Prague to win a Fake Award.
    Prikaži ovu nit
    1 korisnik označava da mu se sviđa
    Poništi
  11. 31. sij

    4) __most importantly__, a "stupid shit i keep forgetting" doc which includes code snippets (10 to 20 lines) i end up reusing. organised by library/tech. occasionally i go through and demote some based on old tech. i would officially be technically useless without this doc! 5/5

    15 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  12. 31. sij

    3) a chats doc; everytime i chat to someone of interest i write a couple of dot points about the conversation. ( if we're very had an interesting video chat or met over coffee in the last 10 years; you are in this doc!!! ) 4/5

    1 reply 7 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  13. 31. sij

    2) a very verbose reproduction journal per project; basically commands i ran, git SHAs, things that worked / didn't, screen caps of graphs etc. i've used these to repro results from years in the past. these docs are linked to from the master journal. 3/5

    1 reply 5 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  14. 31. sij

    1) a master journal with what i worked on today; basically ~5 dot points / day. if i do more than 30min work on something, i add to this doc. have these going back electronically to 2011 and in paper form (A5 hardback) to ~1998. 2/5

    1 proslijeđeni tweet 14 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  15. 31. sij

    love "An Opinionated Guide to ML Research" !! i'm a big big fan of journals! the ones i keep are.... 1/5

    21 proslijeđeni tweet 117 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  16. 31. sij

    reading the docs for jax.lax.scan and had a major theano flashback!! (but not in a bad way)

    1 reply 1 proslijeđeni tweet 13 korisnika označava da im se sviđa
    Poništi
  17. 31. sij

    our train conductor just got a call. his ring tone? thomas the tank engine. haha. classic!!

    1 reply 19 korisnika označava da im se sviđa
    Poništi
  18. 30. sij

    Awesome search auto complete... given "stand alone self attention" completes as "stand alone self attention lyrics". Also, free idea, GAN to sing you papers from arxiv

    4 korisnika označavaju da im se sviđa
    Poništi
  19. proslijedio/la je Tweet
    30. sij

    🔥 OMG Watch how fast a bushfire in can spread

    This Is How Fast A Bushfire Can Spread In 3 Minutes
    Watch how far a bushfire can spread in just 3 MINUTES
    120 proslijeđenih tweetova 147 korisnika označava da im se sviđa
    Poništi
  20. 30. sij

    Yes, thanks! Sorry my original tweet was unclear; I meant w.r.t using a LR finder at the same time. Have used hyper param optimisation a fair bit; keras tuner, vizier, spearmint, hyperopt. My fav at the moment is FB Ax.

    Odgovor korisniku/ci @mat_kelcey
    This was posted today on Tensor Flow Blog, take a look on it. Keras Tuner.
    1 reply 4 korisnika označavaju da im se sviđa
    Poništi
  21. 30. sij

    Todo: write talk

    A product-centric view of machine learning with ML experts and Matt Travers. Register to in on 3 Mar:
    6 korisnika označava da im se sviđa
    Poništi

@mat_kelcey 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:

    ·