Leo Dirac

@leopd

Free-roaming deep learner with a legacy in physics. Former Amazon, Google. he/him

Seattle, WA
Vrijeme pridruživanja: travanj 2007.

Tweetovi

Blokirali ste korisnika/cu @leopd

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

  1. prije 7 sati

    Or Equifax - why do they still exist? Their raison d'être is to hold data, about the most sensitive financial data possible. They leaked most of that, and their punishment was a wrist-slap. They should have been sued to oblivion. But there's no legal basis to do so.

    Prikaži ovu nit
    Poništi
  2. prije 7 sati

    I'm thinking of the recent Google Photos export bug where they handed out videos from other people's accounts. Totally preventable if companies bother to invest in taking privacy seriously. But why bother? That kind of effort costs money.

    Prikaži ovu nit
    Poništi
  3. prije 7 sati

    Clayton Christensen (Innovator's Dilemma) developed insight into "disruptions" that is super powerful and general. If you don't know the difference between what he meant when he coined the term "disruption" and how TechCrunch typically uses it, go read his stuff.

    Poništi
  4. prije 7 sati

    Tech companies should see private customer data as both an asset and a liability. Today there are negligible consequences for data leaks, so companies' only motivation to actually protect customer data is to prevent minor PR hassles. Legislation could fix this.

    Prikaži ovu nit
    Poništi
  5. 25. sij

    Amazing true story about how a forecasting model played a key role in almost starting a nuclear war:

    Poništi
  6. 23. sij

    Set a new personal record (PR) for running up the Blaine steps (150' elevation) of 81.6 seconds. Coincidentally my PR best time to solve the 3x3 Rubik's cube is 87.6 seconds (WCA "average" of 5). I think if I tried to do them both simultaneously, it would be slower than the sum.

    Poništi
  7. proslijedio/la je Tweet
    23. sij
    Poništi
  8. proslijedio/la je Tweet
    22. sij

    We have a paper as well as a joint workshop with accepted at Please, come talk to us about your reward functions! Yunzhe Tao, Saurabh Gupta, and our amazing engineering team!

    Poništi
  9. 22. sij

    Why do supernova remnants appear anisotropic? That is, why do we see discrete rings instead of just spheres? e.g. this image of Supernova 1987a after 20 years of expansion. Is this just gravitational lensing or something like that?

    Poništi
  10. proslijedio/la je Tweet
    21. sij

    We've added more tickets for our meetup next week. Get a ticket at to hear Fardin Abdi from Uber talk about Horovod, and talk about configuration space.

    Poništi
  11. proslijedio/la je Tweet
    19. sij
    Odgovor korisniku/ci
    Poništi
  12. 19. sij

    Today I ran been every Dick’s Drive-In in Seattle, eating at every one. with and others. 5 stops, 18 miles, three milkshakes, three fries, three deluxe and a special. (Minus beef.) Still kinda hungry.

    Poništi
  13. 17. sij

    My judgy watch says my fitness is declining, and knows me well enough to advise more low-aerobic workouts. While explaining its nagging to my kid, I predicted that when she's an adult she'll have a judgy earring that nags her to eat more vegetables. I like working on that tech.

    Poništi
  14. proslijedio/la je Tweet
    16. sij

    Wow! Our upcoming talk sure sold out fast. Those of you still interested in attending, stay tuned. We're figuring out if we have room to issue any more tickets. You'll hear it here first.

    Poništi
  15. 15. sij

    I'm looking forward to diving into details about Horovod in this talk. taught me the basics of distributed deep learning and was been a big proponent of ring allreduce. Cool to see it with wide adoption.

    Poništi
  16. proslijedio/la je Tweet
    14. sij

    If you are interested in working with doodles and sketches rather than just pixel photos, check out this survey of Deep Learning for Free-Hand Sketch, by Xu et al., with 240 references to what used to be a niche area of machine learning and computer vision

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

    Our next event will be January 29 hosted by who will explain how they built Horovod to scale deep learning training jobs to hundreds of GPUs. Get your tickets at

    Prikaži ovu nit
    Poništi
  18. 13. sij

    Now that we understand deep learning reasonably well, it means any process or system that can be differentiated through can be directly optimized with neural networks. Making a differentiable physics simulator is a difficult task, but would enable some amazing things.

    Poništi
  19. proslijedio/la je Tweet

    A short summary of 10 of the decade. It has been an action-packed decade. Thankful for all opportunities I have been given and hoping for more in coming decade.

    Poništi
  20. 28. pro 2019.
    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:

    ·