Tu Vu

@tuvuumass

Ph.D. student at University of Massachusetts Amherst, working on Deep Learning and Natural Language Processing.

Vrijeme pridruživanja: travanj 2017.

Tweetovi

Blokirali ste korisnika/cu @tuvuumass

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

  1. Prikvačeni tweet
    28. srp 2019.

    Excited to share our paper () which improves paragraph classification by pretraining the encoder on unlabeled data using our sentence content objective. Work done with my advisor . Code: . Summary below [1/5]

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

    Break it down! Introducing the "Break" benchmark for testing the ability of models to break down a question into required steps for computing the answer. Accepted for . Learn more about Break in this post by Tomer Wolfson on the AI2 Blog:

    Poništi
  3. proslijedio/la je Tweet
    30. sij

    Humans learn from curriculum since birth. We can learn complicated math problems because we have accumulated enough prior knowledge. This could be true for training a ML/RL model as well. Let see how curriculum can help an RL agent learn:

    Prikaži ovu nit
    Poništi
  4. proslijedio/la je Tweet

    This week's LTI Colloquium is from on "Towards Story Generation"! Come learn how to train models that can efficiently generate more coherent and stylistically consistent stories:

    Poništi
  5. proslijedio/la je Tweet
    29. sij

    New blog post: Contrastive Self-Supervised Learning. Contrastive methods learn representations by encoding what makes two things similar or different. I find them very promising and go over some recent works such as DIM, CPC, AMDIM, CMC, MoCo etc.

    Poništi
  6. 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
  7. proslijedio/la je Tweet
    28. sij

    Introducing , a 2.6B-param open-domain chatbot with near-human quality. Remarkably, we show strong correlation between perplexity & humanlikeness! Paper: Sample conversations:

    Prikaži ovu nit
    Poništi
  8. proslijedio/la je Tweet
    27. sij

    [1/2] Excited to present SMART: Semi-Autoregressive Training for Conditional Masked Language Models. SMART closes the performance gap between semi- and fully-autoregressive MT models, while retaining the benefits of fast parallel decoding. With

    Prikaži ovu nit
    Poništi
  9. proslijedio/la je Tweet
    9. sij

    We ( Ananth, , , ) , are pleased to announce the release of ORB, an Open Reading Benchmark. This is an evaluation server that tests a single model on a variety of reading comprehension datasets (SQuAD, DROP, Quoref, ...).

    Prikaži ovu nit
    Poništi
  10. proslijedio/la je Tweet
    8. sij

    Excited that our work on characterizing racial bias in football commentary was covered by ! Thanks to for thoroughly explaining and contextualizing our research to a non-academic audience. We're continuing with this work, so look out for more soon!

    Poništi
  11. proslijedio/la je Tweet
    1. sij

    We present our new year special: “oLMpics - On what Language Model pre-training captures״, , Exploring what symbolic reasoning skills are learned from an LM objective. We introduce 8 oLMpic games and controls for disentangling pre-training from fine-tuning.

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

    Our work on nearest neighbor language models has been accepted to Woohoo!! Code coming in the new year!

    Poništi
  13. 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
  14. proslijedio/la je Tweet
    12. pro 2019.

    Our MixTape is 3.5-10.5x faster than Mixture of Softmaxes /w SOTA results in language modeling & translation. Key is to do gating in the logit space but with vectors instead of scalars (+sigmoid tree decomposition & gate sharing for efficiency). /w Zhilin, Russ, Quoc

    Poništi
  15. proslijedio/la je Tweet
    3. pro 2019.

    Ever wanted to combine the NLU superpowers of BERT with the generation superpowers of GPT-2? It's now possible in transformers thanks to !

    Poništi
  16. proslijedio/la je Tweet
    26. stu 2019.

    Transformers v2.2 is out, with *4* new models and seq2seq capabilities! ALBERT is released alongside CamemBERT, implemented by the authors, DistilRoBERTa (twice as fast as RoBERTa-base!) and GPT-2 XL! Encoder-decoder with ⭐Model2Model⭐ Available on

    Poništi
  17. proslijedio/la je Tweet
    19. stu 2019.

    New blog post! Lots of BERT compression papers lately... I put them all in one place and did a brief taxonomy.

    Poništi
  18. proslijedio/la je Tweet
    12. stu 2019.

    Another view of Noisy Student: semi-supervised learning is great even when labeled data is plentiful! 130M unlabeled images yields 1% gain over previous ImageNet SOTA that uses 3.5B weakly labeled examples! joint work /w , Ed Hovy,

    Poništi
  19. proslijedio/la je Tweet
    12. stu 2019.

    Want to improve accuracy and robustness of your model? Use unlabeled data! Our new work uses self-training on unlabeled data to achieve 87.4% top-1 on ImageNet, 1% better than SOTA. Huge gains are seen on harder benchmarks (ImageNet-A, C and P). Link:

    Prikaži ovu nit
    Poništi
  20. proslijedio/la je Tweet
    11. stu 2019.

    Excited to host this Thursday for a talk on "Rethinking Transformers for machine translation and story generation":

    Poništi
  21. proslijedio/la je Tweet
    10. stu 2019.

    😊Self-supervised learning opens up a huge opportunity for better utilizing unlabelled data while learning in a supervised learning manner. My latest post covers many interesting ideas of self-supervised learning tasks on images, videos & control problems:

    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:

    ·