Ryan Foo

@ryanmyfoo

(Tiny)ML research, voice, software, games, blockchain, VC. . Write at (he/him) To boldly recognize voices

Singapore
Vrijeme pridruživanja: studeni 2018.

Tweetovi

Blokirali ste korisnika/cu @ryanmyfoo

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

  1. Prikvačeni tweet
    31. pro 2019.

    Singapore going absolutely wild with the new year’s fireworks. DRONES. Drones.

    Prikaži ovu nit
    Poništi
  2. proslijedio/la je Tweet

    To the person who figured out my honeypot is a honeypot could you please stop putting the picture of Pooh bear with a jar of honey on it? Its like this person's life mission, I've blocked him on: - Client - IPs (now on Tor ffs) - The image (he just edits 1 pixel every time...)

    Prikaži ovu nit
    Poništi
  3. prije 2 sata

    reading 's brain food newsletter. his thought: as you get better at problem solving, you move to avoid problems before they start. this is low visibility. for programming: instead of programming yourself into a frenzy, think about architecture... Wu Wei

    Poništi
  4. prije 2 sata

    so I found dude_miaowing.wav in the _background_noises_ folder in 's dataset and my background noise has never been the same ever since...

    Poništi
  5. prije 4 sata

    currying but with arrows

    Poništi
  6. prije 19 sati

    "So I used code and algorithms in order to achieve this task..."

    Poništi
  7. prije 20 sati

    just me, casually opening emacs via the command line. I'm a GNU emacs on Linux kind of guy. emacs somefile.txt. some file.

    Poništi
  8. 3. velj

    yes, you are right -- I do enjoy a good delta compression from time to time.

    Poništi
  9. proslijedio/la je Tweet

    “I always use my clients’ products [...] and why not, pray tell? Are these not the best goods and services on earth? I think they are, and that is why I advertise them.” I admire the way he realized his vision of making an excellent, joyful living w/o compromising his values

    Prikaži ovu nit
    Poništi
  10. proslijedio/la je Tweet
    1. velj

    Reasons why startups should price relatively high in general: 1. Easier to lower prices than raise them 2. PMF at low price doesn't mean PMF at a higher one 3. Less likely to leave money on the table 4. If your product is actually 10x better, it's worth it What else?

    Poništi
  11. 1. velj

    think big...

    Poništi
  12. 31. sij

    Ok maybe after more substantive example has been implemented

    Prikaži ovu nit
    Poništi
  13. 31. sij

    Wow, really lost to Berkeley Overmind in 2010 as a PhD student in Computer Science, 8 years later led the team that built 's AlphaStar. I am WEEPING.

    Poništi
  14. 31. sij

    I'll be right back, making a pull request to library wish me luck team

    Prikaži ovu nit
    Poništi
  15. 31. sij

    My Jan 2020; Started at GovTech, machine learning research. Learnt a lot about signal processing and audio. Won Hack and Roll 2020 w great team. Youth Corps project starting up. CNY steamboat w family. Started climbing. Worked on Javascript. Duolingo Russian streak.

    Poništi
  16. 31. sij

    But perhaps I am being myopic here. Neural nets died out twice in the 60s and 80s, winters. Why stare so hard at the next few years, when there’s a marathon to run?

    Poništi
  17. 31. sij

    But I guess it’s like lamenting the single layer perceptron. Or TD-Learning. They had their day in the sun. Now they are relics of history in practice, only used to tell stories to neophytes about why they should study attention. (And so this is better because...)

    Prikaži ovu nit
    Poništi
  18. 31. sij

    , when we’re looking at speech, a common approach has been to convert the signal to a spectrogram (2D image representing a sound wave) before training a convolutional net on the spectrogram. Fast. Fourier. Transform.

    Prikaži ovu nit
    Poništi
  19. 31. sij

    Studied recurrent nets for my final research project only to hear them get replaced almost instantly-feels bad man. Computations being done in sequence == lots of waiting and wasting of power.Attention models can be parallelized. So can ConvNets. So even for sequence processing/1

    Prikaži ovu nit
    Poništi
  20. 31. sij

    reviving this meme if(goingToCrashIntoEachOther) { dont(); }

    Poništi
  21. 30. sij

    brain feeling overstimulated, I really want to talk about transformers and attention w someone right now to understand it better

    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:

    ·