noe casas

@noecasas

Neural Machine Translation PhD student and researcher at Lucy Software. Former data scientist and sw engineer. Chinese lang enthusiast. Life-long learner. 学无止境

Vrijeme pridruživanja: svibanj 2009.

Tweetovi

Blokirali ste korisnika/cu @noecasas

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

  1. proslijedio/la je Tweet
    31. sij

    I am a doctor now!! 🎓🎉🎊🎓

    Poništi
  2. proslijedio/la je Tweet
    30. lis 2019.

    An arXiv overlay journal is one where the articles are not hosted by the journal itself but simply live on arXiv. That doesn't mean they are "just preprints" -- they are refereed in the usual way and are prepared using a style file.

    Prikaži ovu nit
    Poništi
  3. 19. ruj 2019.

    Using fairseq and finding it difficult to debug your model because pdb does not work with multiprocessing? Suffer no more and discover fairseq's pdb wrapper: "from fairseq import pdb; pdb.set_trace()" . Find more at

    Poništi
  4. proslijedio/la je Tweet
    15. ruj 2019.

    The 3rd edition of our MSc course on Deep Learning for Artificial Intelligence kicks off this Monday. Once again, an extraordinary amount of registered students (52), a 100% growth since 2017.

    , , i još njih 4
    Poništi
  5. proslijedio/la je Tweet
    4. ruj 2019.

    It was great working on this project during my internship at , and our paper has now been accepted at ! Paper: Code:

    Tweet je nedostupan.
    Poništi
  6. proslijedio/la je Tweet
    5. kol 2019.
    Poništi
  7. 2. kol 2019.

    Are you doing NMT and are curious about what interesting things were presented at ? Check out my summary at

    Poništi
  8. proslijedio/la je Tweet

    In Latent Structure Models tutorial, we showed a slightly diff. justification for Straight-Through and SPIGOT as adding pseudo-supervision w/ perceptron loss btw. latent var & a "pulled-back" label, induced from downstream loss. Tech note:

    Prikaži ovu nit
    Poništi
  9. proslijedio/la je Tweet
    28. srp 2019.

    kicking off our tutorial on unsupervised cross-lingual learning Slides:

    Poništi
  10. proslijedio/la je Tweet
    24. srp 2019.

    If you are Barcelona and want to dive into deep learning, you can already apply to our undergraduate school between 22-28 January 2020. Third edition 🤟 Instructors: @ramonmorros & myself

    Poništi
  11. proslijedio/la je Tweet
    17. srp 2019.

    If you are Barcelona, you should sign up for our MSc course on Deep Learning for AI. 3rd edition, better, stronger and, obviously, deeper than ever. Welcoming from as instructor 😀

    , , i još njih 6
    Poništi
  12. proslijedio/la je Tweet
    4. srp 2019.

    imho didn't receive the attention it should have: 1st-order msg: If you are using 32k BPE tokens for Transformer you can't claim "a strong baseline". 2nd-order: you can't carry hyperparamters best practice across models (LSTM -> Transformer)

    Poništi
  13. proslijedio/la je Tweet
    4. srp 2019.

    RL methods for MT (MRT, GANs and REINFORCE) might not get their performance boost from actually improving translations. Moreover, there are no convergence guarantees for MRT. For these practical and theoretical (first for MRT?) results. Our new Preprint

    Prikaži ovu nit
    Poništi
  14. proslijedio/la je Tweet
    7. lip 2019.

    ❗El será sede de uno de los grandes superordenadores europeos. La UE podría aportar cerca de 100 millones€ para MareNostrum 5, la inversión más alta en una infraestructura de investigación en España

    Prikaži ovu nit
    Poništi
  15. proslijedio/la je Tweet
    23. svi 2019.

    So proud of and team for launching GitHub Sponsors: a first step in a long journey towards providing financial infrastructure for developers.

    Poništi
  16. proslijedio/la je Tweet
    23. svi 2019.

    Happy to (finally) share SciHive - arXiv on steroids 🚀 ✓ Rank trending papers by Twitter ✓ Comment and ask questions on papers ✓ Collaborate within a group ✓ Save private notes ✓ View table of contents ✓ Preview references ✓ Open source

    Poništi
  17. 28. tra 2019.
    Poništi
  18. proslijedio/la je Tweet
    16. tra 2019.

    1/ Let me share my "Deep Architectures" slides, where I make a review of different basic architectures (FC, CNN, RNN/LSTM) and their compositions (QRNNs, AEs, Skip connections, Deep classifiers, GANs, ...):

    Prikaži ovu nit
    Poništi
  19. proslijedio/la je Tweet

    La mayoría de jóvenes investigador@s trabajamos en condiciones de precariedad laboral. Que los gobiernos maquillen un presupuesto con triquiñuelas contables y préstamos irrealizables mientras otras pasamos penurias para investigar es un insulto.

    Poništi
  20. proslijedio/la je Tweet
    21. ožu 2019.

    Excited to hold this course again in beautiful San Sebastian, this time with an invited talk by Please inform anyone that might be interested

    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:

    ·