Shijie Wu

@EzraWu

Ph.D. student at . Previous intern (LATTE). He/Him.

Baltimore, MD
Joined December 2008
19 Photos and videos Photos and videos

Tweets

You blocked @EzraWu

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

  1. 19 Nov 2021

    The thesis and slides will be available online shortly after the finalization of the thesis and one under review chapter getting cleared

    8 likes
    Show this thread
    Show this thread
    Undo
  2. 18 Nov 2021

    Life update: I have successfully defended my thesis and will join Bloomberg AI in January

    44 replies 28 retweets 1,606 likes
    Show this thread
    Show this thread
    Undo
  3. 15 Sep 2021

    Resources we released: * Large English-Arabic encoder * Data projection * Arabic denormalization toolkit [4/4]

    1 retweet 5 likes
    Show this thread
    Show this thread
    Undo
  4. 15 Sep 2021

    We also conducted multilingual experiments with 8 target languages. This is a joint work with Mahsa Yarmohammadi Jialiang Guo [3/4]

    1 reply 1 retweet 1 like
    Show this thread
    Show this thread
    Undo
  5. 15 Sep 2021

    We conducted extensive experiments on EN-->AR as test case including: 1️⃣ data projection pipeline with various MT models and aligners; 2️⃣ impact of encoder on MT and aligner; 3️⃣ impact of fine-tuning on aligner; 4️⃣ self-training as controlled comparison of data projection. [2/4]

    1 reply 1 like
    Show this thread
    Show this thread
    Undo
  6. 15 Sep 2021

    Q: What's the best choice to improve zero-shot cross-lingual transfer performance if u can’t manually annotate any data? A: The best setup is task dependent, so try multiple setups! 📈 Check out our paper “Everything Is All It Takes” [1/4]

    3 replies 13 retweets 41 likes
    Show this thread
    Show this thread
    Undo
  7. Retweeted
    6 Aug 2021

    🚀We are excited to share new languages! As a part of shared task on morphological inflection, we added data for ✨32 languages from 13 families✨. We analyzed systems' predictions on them, conducting an extensive error analysis: 1/8

    1 reply 18 retweets 42 likes
    Show this thread
    Show this thread
    Undo
  8. 25 Apr 2021

    Happy to share that our paper on "Applying the Transformer to Character-level Transduction" won the honorable mention for short paper awards! Congrats and Mans! Talk Paper Code

    8 retweets 49 likes
    Undo
  9. Retweeted
    16 Feb 2021

    "Differentiable Generative Phonology", in collaboration with and , is finally out! Tired: Asking linguists to posit discrete underlying forms Wired: learning continuous underlying forms end-to-end

    1 reply 12 retweets 37 likes
    Show this thread
    Show this thread
    Undo
  10. Retweeted
    17 Nov 2020

    Which *BERT (and how can we improve *BERT-science)? Come to the Zoom QA session 13 at 11am-12 EST. See thread for a tl;dr: And chat about constant memory coreference resolution at Gather session 5, tomorrow 1-3pm EST:

    Which *BERT should you use? We (with and ) surveyed as many of them as we could in this paper . Takeaways: 1) Encoders can be expensive to use and may not be worth it if they only get you +0.2% ... (1/3)
    Show this thread
    1 retweet 7 likes
    Undo
  11. 17 Nov 2020

    Summary of "Which *BERT? A Survey Organizing Contextualized Encoders"

    Which *BERT should you use? We (with and ) surveyed as many of them as we could in this paper . Takeaways: 1) Encoders can be expensive to use and may not be worth it if they only get you +0.2% ... (1/3)
    Show this thread
    2 likes
    Show this thread
    Show this thread
    Undo
  12. 17 Nov 2020

    Summary of "Do Explicit Alignments Robustly Improve Multilingual Encoders?"

    Happy to share our paper (w/ ) on contrastive cross-lingual alignment titled “Do Explicit Alignments Robustly Improve Multilingual Encoders?” Expand thread for TL;DR ... (1/3) Paper Code
    Show this thread
    1 reply
    Show this thread
    Show this thread
    Undo
  13. 17 Nov 2020

    Q: Do Explicit Alignments Robustly Improve Multilingual Encoders? My talk: Code: Meeting: Gather town in 10 mins Q: Which *BERT should I use? 's talk: Meeting: Zoom, Nov 18, 11-12 EST

    Happy to share that one short paper "Do explicit alignments robustly improve massively multilingual encoders?" w/ and one long paper "Which *BERT? A Survey Organizing Contextualized Encoders" w/ Patrick Xia and got accepted to (1/2)
    Show this thread
    1 reply 1 retweet 5 likes
    Show this thread
    Show this thread
    Undo
  14. Retweeted
    28 Oct 2020

    DATASET RELEASE: "CC100", the CommonCrawl dataset of 2.5TB of clean unsupervised text from 100 languages (used to train XLM-R) is now publicly available. You can find below the Data: Script: By @VishravC et al.

    3 replies 199 retweets 708 likes
    Undo
  15. Retweeted
    2 Nov 2020

    Ok here we go. Look at the electoral maps by county for the last few decades of US presidential elections.  You’ll notice that the South goes almost uniformly Republican red every time. Duh. But if you look closer, there’s something else there ...

    505 replies 16,840 retweets 51,114 likes
    Show this thread
    Show this thread
    Undo
  16. 20 Oct 2020

    Interesting work showing word translation from mBERT with template: "The word ‘SOURCE’ in LANGUAGE is: [MASK]."

    Does multilingual BERT know how to translate? Can we neutralize language from its representations? Check out our new paper: "It’s not Greek to mBERT: Inducing Word-Level Translations from Multilingual BERT", joint work with  and  (1/4)
    Show this thread
    5 likes
    Undo
  17. Retweeted
    15 Oct 2020

    New EMNLP paper with and -- With Little Power Comes Great Responsibility -- (1/3)

    4 replies 35 retweets 150 likes
    Show this thread
    Show this thread
    Undo
  18. 9 Oct 2020

    … (3) quality of bitext has small impact on downstream performance (4) while alignment helps mBERT in some cases, none of the methods we considered improve XLMR (5) a bigger model leads to much bigger gain compared to ad-hoc alignment, without any need for bitext (3/3)

    1 like
    Show this thread
    Show this thread
    Undo
  19. 9 Oct 2020

    ... (1) contrastive alignment consistently outperforms L2 alignment and performs more robustly than linear mapping (2) zero/few-shot cross-lingual transfer (includ. benchmarks) needs to report mean & variance w/ different seeds, similar to the few-shot learning community .. (2/3)

    1 reply 1 like
    Show this thread
    Show this thread
    Undo
  20. 9 Oct 2020

    Happy to share our paper (w/ ) on contrastive cross-lingual alignment titled “Do Explicit Alignments Robustly Improve Multilingual Encoders?” Expand thread for TL;DR ... (1/3) Paper Code

    1 reply 5 retweets 39 likes
    Show this thread
    Show this thread
    Undo

@EzraWu 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

    ·