gdupont

@ggdupont

程序猿 🐒 doing data science at AIRBUS AI Research - All opinions are obviously my own.

Vrijeme pridruživanja: kolovoz 2010.

Tweetovi

Blokirali ste korisnika/cu @ggdupont

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

  1. proslijedio/la je Tweet
    prije 19 sati

    GPT2 (via 🤗) can now be used & fine-tuned in ParlAI!🦜 e.g. talk to it with: python examples/interactive.py -m hugging_face/gpt2 --gpt2-size large --inference topk --topk 10 Code:

    Poništi
  2. proslijedio/la je Tweet

    So what are the functions of language? On top of my head: - Language as communication channel (receive and convey information about the world) - Language as affordance (saying something = action towards a goal) - Language as social lubricant (chitchat, etc)

    Prikaži ovu nit
    Poništi
  3. proslijedio/la je Tweet
    4. velj

    Very happy that our work “Conversations with Documents. An Exploration of Document-Centered Assistance” has been accepted as a full paper at CHIIR! Work done during my internship at MSR /w Rob Sim, Elnaz Nouri, , and Ryen White. Paper:

    Poništi
  4. proslijedio/la je Tweet
    3. velj

    I am a Wuhan girl living in Hong Kong. Right now, all my close family is under lock-down back in my hometown, at the epicenter of this epidemic.

    Prikaži ovu nit
    Poništi
  5. proslijedio/la je Tweet
    3. velj
    Poništi
  6. proslijedio/la je Tweet
    3. velj

    is more than just public code. It's a mindset of sharing, being transparent and collaborating across organizations. It's about building on the shoulders of other projects and advancing together the state of technology (1/N) , , ,

    Prikaži ovu nit
    Poništi
  7. 3. velj

    I tend to think that going full speed on "a model of the world" is super hard... thus the use with open world chatbots. Focussing on a specific domain allows to restrict the "world" to b taken into account and focus on the language and message passing strategy

    Poništi
  8. proslijedio/la je Tweet

    If an algorithm has "mastered natural language", I would expect it to be able to do some of the things language is for -- communicating information, receiving information, acting on the world... Not merely output something that statistically sounds like language.

    Prikaži ovu nit
    Poništi
  9. proslijedio/la je Tweet

    It’s truly impressive China. Huoshenshan hospital in Hubei it’s not only been built in 10 days, but it’s covered by 5G and a new cloud base medical logging and recording system. It will be running at its full capacity (1000beds) starts from Feb 4th.

    Poništi
  10. proslijedio/la je Tweet
    7. sij

    . people’s papers #2—ELECTRA: and colleagues (incl. at ) show how to build a much more compute/energy efficient discriminative pre-trainer for text encoding than BERT etc. using instead replaced token detection

    Poništi
  11. proslijedio/la je Tweet
    29. sij

    Evolutionary computation used in designing neural networks for natural language processing.

    Poništi
  12. 2. velj

    Always good to see open data shared especially for health issues. Just a gentle reminder that each line in this table is a person and I hope anyone can process this dat with care and following privacy preserving methods.

    Poništi
  13. proslijedio/la je Tweet
    1. velj

    Very happy to share our new preprint! Animals have a remarkable capacity to learn new motor skills, but how does learning change neural population dynamics underlying movement?

    Prikaži ovu nit
    Poništi
  14. proslijedio/la je Tweet
    31. sij

    WinoGrande, a new dataset by researchers on AI2's team, was created in pursuit of generalizable . 's covers the challenges of building that can truly understand human language:

    Prikaži ovu nit
    Poništi
  15. proslijedio/la je Tweet
    1. velj
    Odgovor korisniku/ci
    Poništi
  16. proslijedio/la je Tweet
    1. velj

    Until last month, the only thing I really knew about Wuhan was that it had a rocking live music scene (because we published like a dozen features on Wuhan punk, Wuhan black metal, Wuhan hip-hop etc). Now it'll be forever associated with . I know there's worse things

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

    Curriculum for Reinforcement Learning "Learning is probably the best superpower we humans have." explores four types of curricula that have been used to help RL models learn to solve complicated tasks.

    Poništi
  18. proslijedio/la je Tweet
    1. velj

    Rotation, Translation, and Cropping for Zero-Shot Generalization Makes a lot of sense. Try playing Doom not from an agent-centric perspective! I think agent-centric view is a better prior for encoding useful information using fewer bits for the policy.

    Poništi
  19. proslijedio/la je Tweet
    1. velj

    To the left, you see a trained agent playing a level of a game. To the right, you see the same playthrough from an agent-centric perspective: cropped, translated, and rotated with the agent in the center. Which perspective is the best input for the agent?

    Prikaži ovu nit
    Poništi
  20. proslijedio/la je Tweet
    31. sij

    Only a few more hours till today's SEA! We're looking forward to hosting Alexandra Olteanu (, MSR NY) on problematic text suggestions and Svitlana Vakulenko (, ILPS Amsterdam) on QA over knowledge graphs. Reserve your spot here:

    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:

    ·