François CholletOvjeren akaunt

@fchollet

Deep learning . Creator of Keras, neural networks library. Author of 'Deep Learning with Python'. Opinions are my own.

Mountain View, CA
Vrijeme pridruživanja: kolovoz 2009.

Medijski sadržaj

  1. PyPI downloads for TensorFlow (and its closest competitor, added for scale). Notice how it starts jumping after the release of TF 2.0 late last year (the short gap afterwards is the holiday break) Up and to the right 📈

  2. Much like an airplane, language has a function. No matter how realistic you make your image of a plane, if it misses the function, you still haven't mastered human flight.

    Prikaži ovu nit
  3. Refreshingly honest

  4. I don't get why I am forced by the app to give a wide public platform to these guys. Being away from Twitter was psychologically freeing -- no daily insults, no constant influx of gross ignorance and stupidity and dudes being mad at what they imagine other people are like.

    Prikaži ovu nit
  5. TensorFlow has a suite of tools for optimizing your models for faster inference: This includes post-training weight quantization, and gradual weight pruning during training for your Keras models.

  6. NIPS 1990 had papers about "hybrid symbolic-connectionist" models. This is now as far in the past as 2050 is in the future.

    Prikaži ovu nit
  7. Take care of yourself, and work purposefully

    Prikaži ovu nit
  8. Deep learning refers to an approach to representation learning where your model is a chain of modules (typically a stack / pyramid, hence the notion of depth), each of which could serve as a standalone feature extractor if trained as such. That's also how I define it in my book.

    Prikaži ovu nit
  9. A clean TensorFlow 2 implementation of StyleGAN 2 (with pretrained weights for generating landscapes)

  10. Saturday infosec advice: don't patch, dare to live dangerously

  11. Started reading through this book (pictured here with an edge TPU and my new assistant). It's a great resource for production deep learning best practices -- covers TF Lite, TF.js, CoreML, and more.

  12. Built-in losses and metrics in Keras follow the signature `loss(y_true, y_pred, sample_weight=None)`. If you have exotic losses or metrics, a simple way to add them w/o having to implement your own training loop from scratch is to define them in an "endpoint layer". Like this:

    Prikaži ovu nit
  13. The view from the cyberwindow at the cyberbar where I'm currently sipping cyberwhisky

  14. 1981 DeLorean, 1985 Lotus Esprit

    Prikaži ovu nit
  15. The toy version of the Dome Zero, which is basically the cybertruck

    Prikaži ovu nit
  16. The Dome Zero, 1970s Japanese supercar that never made it to production

    Prikaži ovu nit
  17. Total Recall (1990)

    Prikaži ovu nit
  18. Love the new Tesla cybertruck (design by , for the Ghost in the Shell movie that had Scarlett Johansson in it)

    Prikaži ovu nit
  19. The new TextVectorization layer in Keras enables you to put your text preprocessing directly inside your models, in a way that is automatically serializable and deployable. You won't have to worry about separately keeping track of your vocabulary again. Here's how to use it:

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:

    ·