Guillaume Chevalier

@guillaume_che

Passionned about the applications of . Loves to solve complicated and to innovate creatively.

Canada
neuraxio.com
Joined September 2016
13 Photos and videos Photos and videos

Tweets

You blocked @guillaume_che

Are you sure you want to view these Tweets? Viewing Tweets won't unblock @guillaume_che

  1. Pinned Tweet
    May 2

    You can now "pip install conv", my first package uploaded on : a tiny library for doing clean and concise convolutional loops. Do neat convolution-styled loops on your lists. 1D and 2D convolutions are supported.

    7 likes
    Undo
  2. Aug 16

    My first solo paper on arXiv! The LARNN: Linear Attention Recurrent Neural Network. Yes, it LARNNs. (Okay, I wrote that in May, but anyway here it is on arXiv)

    1 reply 2 retweets 17 likes
    Undo
  3. Retweeted
    Aug 1

    Pretty mind-blowing paper, 3D-printed representations of neural networks you can run inference on by shining light through... literal *light speed* inference times. They only have classification networks so far, but I'd love to see pix2pix or a GAN soon.

    10 replies 348 retweets 796 likes
    Show this thread
    Show this thread
    Undo
  4. Jul 7

    I didn't expect this question to cause a buzz on StackExchange haha!

    How to COMPLETELY turn off system beep forever for good for real
    Undo
  5. Jul 2

    Just thought of a U-Net RNN. Wouldn't that yield awesome results on various tasks? I've been thinking a lot on transforming signal from one format to another format, and a U-Net RNN sounds promising.

    Undo
  6. Jun 19

    The website is alive! May contain traces of neurons.
    1 like
    Undo
  7. Jun 14

    2 likes
    Undo
  8. Jun 12

    This is awesome.

    What I've been working on for the past year! Inspired by CoVE, ELMo, and ULMFiT we show that a single transformer language model can be finetuned to a wide variety of NLP tasks and performs very well with little tuning/tweaking.
    1 like
    Undo
  9. May 26

    Launching and setting up my Deep Learning consulting company, ! 😀

    Hello Twitter!
    2 replies 1 retweet 10 likes
    Undo
  10. May 20

    7/ The fact that the Multi-Head Attention Mechanisms might work better with added activations on its linear layers (at least in this context) is an interesting discovery. See figure, where test accuracy over time is shown.

    Show this thread
    Show this thread
    Undo
  11. May 16

    And here is the Annotated Multi-Head Attention Mechanism, an analysis of mine: Printing the dimensions of the attention mechanism and its positional embedding, plus some experimental modifications!

    The Annotated Transformer: Line-by-Line PyTorch implementation of "Attention is All You Need"
    3 likes
    Undo
  12. May 15

    6/ Here is some code for a concrete implementation of the LARNN I've made: Yes, it LARNNs. It uses a Multi-Head Attention Mechanism as in Attention Is All You Need, and the equations are derived from the LSTM cell.

    1 reply 1 retweet 5 likes
    Show this thread
    Show this thread
    Undo
  13. Retweeted
    May 14

    LARNN: Linear Attention Recurre... by

    1 retweet 3 likes
    Undo
  14. May 8

    This is neat

    I have published an introduction to Einstein summation (einsum) notation for Deep Learning. Einsum is the Swiss Army knife for all kinds of tensor operations. It is available in and with v0.4 now also in .
    1 like
    Undo
  15. Retweeted
    May 5

    It's fascinating to see what increases superlinearly (patents, GDP, electricity, crime, AIDS cases) and sublinearly (gas stations, road surface). And, of course, a priori it's not obvious that any of these should be functions so purely of population.

    11 replies 108 retweets 363 likes
    Show this thread
    Show this thread
    Undo
  16. Retweeted
    May 5

    To all the machine learning researchers who sat on the sidelines during the "chihuahua or muffin?" wars, we need your support now more than ever.

    Ticks can be the size of a poppy seed. Can you spot all 5 ticks in this photo? Learn how to prevent tick bites.
    Show this thread
    1 reply 12 retweets 78 likes
    Undo
  17. Retweeted
    May 2

    We rely on 15 people to do our science. Without them matplotlib, numpy and pandas would not be maintained.

    85 replies 3,855 retweets 6,296 likes
    Show this thread
    Show this thread
    Undo
  18. Retweeted
    Apr 30

    XKCD tackles Python packaging today 😀 

    26 replies 490 retweets 1,224 likes
    Show this thread
    Show this thread
    Undo
  19. May 2

    Haha, this is excellent:

    Undo
  20. Retweeted
    Apr 25

    Yesterday class was more hands-on. We went through the new features introduced by 0.4.0 and analysed the DCGAN training example. Finally, we provided some pointers about the checkerboard artefacts which are nicely fixed following

    , and
    34 retweets 121 likes
    Undo

@guillaume_che hasn't Tweeted yet.

Loading seems to be taking a while.

Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.

    You may also like

    ·