Yung-Sung Chuang

@YungSungChuang

NLP & Speech@National Taiwan University.

Taipei City, Taiwan
Vrijeme pridruživanja: veljača 2019.

Tweetovi

Blokirali ste korisnika/cu @YungSungChuang

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

  1. proslijedio/la je Tweet
    30. sij

    We're standardizing OpenAI's deep learning framework on PyTorch to increase our research productivity at scale on GPUs (and have just released a PyTorch version of Spinning Up in Deep RL):

    Poništi
  2. proslijedio/la je Tweet
    28. sij

    New paper: Towards a Human-like Open-Domain Chatbot. Key takeaways: 1. "Perplexity is all a chatbot needs" ;) 2. We're getting closer to a high-quality chatbot that can chat about anything Paper: Blog:

    Prikaži ovu nit
    Poništi
  3. proslijedio/la je Tweet
    24. sij

    We're releasing mBART, a new seq2seq multilingual pretraining system for machine translation across 25 languages. It gives significant improvements for document-level translation and low-resource languages. Read our paper to learn more:

    Poništi
  4. proslijedio/la je Tweet
    16. sij

    Music you love, powered by AI 🎧🎶 See how the streaming service QQ Music from Tencent uses TensorFlow to manage their extensive music library and enhance user experience! Learn more →

    Poništi
  5. proslijedio/la je Tweet
    10. sij

    Now that neural nets have fast implementations, a bottleneck in pipelines is tokenization: strings➡️model inputs. Welcome 🤗Tokenizers: ultra-fast & versatile tokenization led by : -encode 1GB in 20sec -BPE/byte-level-BPE/WordPiece/SentencePiece... -python/js/rust...

    Prikaži ovu nit
    Poništi
  6. proslijedio/la je Tweet
    9. sij
    Poništi
  7. proslijedio/la je Tweet
    27. kol 2019.

    Do neural networks learn what we think they learn? reviews research that suggests that they often instead fall prey to the so-called Clever Hans effect and discusses its implications for NLP.

    Poništi
  8. 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
  9. proslijedio/la je Tweet
    6. sij

    . people’s papers #1— and colleagues (incl. at ) show the power of neural nets learning a context similarity function for kNN in LM prediction—almost 3 PPL gain on WikiText-103—maybe most useful for domain transfer

    Poništi
  10. proslijedio/la je Tweet
    6. sij

    10 ML & NLP Research Highlights of 2019 New blog post on ten ML and NLP research directions that I found exciting and impactful in 2019.

    Poništi
  11. proslijedio/la je Tweet
    4. sij

    I liked the LSH attention in the reformer Sparse, efficient, simple Dynamic sparse attn is fascinating & mostly dealt by – softmax+topK: Recurrent Independent Mech. (MILA) Product-Key Mem (FB) – 𝛂-entmax: Adap. Sparse Transformer (DeepSPIN) links👇[1/3]

    Prikaži ovu nit
    Poništi
  12. proslijedio/la je Tweet
    1. sij

    It's January 1st, which means... 🎊🎉 we can FINALLY leave Python 2 behind!! 🎉🎊

    Prikaži ovu nit
    Poništi
  13. proslijedio/la je Tweet
    29. pro 2019.

    After getting published in ICLR as an Independent Researcher, I have received nearly 100 messages from others who are looking to do the same. So I wrote a blog post on why I decided to do it and my advice to others.

    Prikaži ovu nit
    Poništi
  14. proslijedio/la je Tweet
    25. pro 2019.
    Prikaži ovu nit
    Poništi
  15. proslijedio/la je Tweet
    20. pro 2019.

    BERT learns syntax and semantics, but what about real-world and common sense knowledge? Our paper proposes a new way to tech BERT about real-world entities. Congrats , and for ICLR acceptance.

    Prikaži ovu nit
    Poništi
  16. proslijedio/la je Tweet
    20. pro 2019.

    UMAP is often said to be superior to t-SNE because it is better at preserving global distances. Dmitry and I showed that this is because by default, t-SNE initializes randomly whereas UMAP initializes with Laplacian eigenmaps.

    Prikaži ovu nit
    Poništi
  17. proslijedio/la je Tweet
    20. pro 2019.

    Yes! I got my first big conference paper accepted at ICLR, with spotlight! We improve the previous DeepMind paper "NALU" by 3x-20x. – This took 7-8 months, working without any funding as an independent researcher. Paper: Code:

    Prikaži ovu nit
    Poništi
  18. proslijedio/la je Tweet
    20. pro 2019.

    We spend our time finetuning models on tasks like text classif, NER or question answering. Yet 🤗Transformers had no simple way to let users try these fine-tuned models. Release 2.3.0 brings Pipelines: thin wrappers around tokenizer + model to ingest/output human-readable data.

    , , i još njih 4
    Prikaži ovu nit
    Poništi
  19. proslijedio/la je Tweet
    20. pro 2019.

    ALBERT is a new, open-source architecture for natural language processing that achieves state-of-the-art performance on multiple benchmarks with ~30% fewer parameters than . Learn all about it below:

    Poništi
  20. proslijedio/la je Tweet
    20. pro 2019.

    Finally can reveal our paper on ELECTRA, much more efficient than existing pretraining, state-of-the-art results; more importantly, trainable with one GPU! Key idea is to have losses on all tokens. Joint work , , .

    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:

    ·