Andrew M. Dai

@iamandrewdai

Deep learning researcher at Google Health

San Francisco, CA
research.google.com/pubs/AndrewDai…
Vrijeme pridruživanja: veljača 2011.
5 fotografija ili videozapisa Fotografije i videozapisi

Tweetovi

Blokirali ste korisnika/cu @iamandrewdai

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

  1. proslijedio/la je Tweet
    13. pro 2019.

    Another talk from our team at My teammates Dale Webster and talking about the challenges in building for real-world use, sharing experience beyond model build/evaluation, into integration in real products, real workflows, for real people

    3 proslijeđena tweeta 28 korisnika označava da im se sviđa
    Prikaži ovu nit
    Prikaži ovu nit
    Poništi
  2. 13. pro 2019.

    Come see our team's work on medical records: doctor's notes, uncertainty, EHR graphical structure, clinical forecasting, federated and differentially private learning and multivariate timeseries at the ML4H workshop at today. Code also at

    3 proslijeđena tweeta 36 korisnika označava da im se sviđa
    Poništi
  3. 10. pro 2019.

    Unexpected NeurIPS moments: main conference poster sessions reaching capacity.

    5 replies 27 proslijeđenih tweetova 160 korisnika označava da im se sviđa
    Poništi
  4. 10. pro 2019.

    Come to the Google booth at NeurIPS today to hear about some of our recent research and open-sourcing on medical records!

    During the lunch break (12:45 - 2:15), drop by the Google booth to learn about our research internship opportunities. You can also get more information on Google’s latest health-related efforts () with researcher .
    1 proslijeđeni tweet 17 korisnika označava da im se sviđa
    Poništi
  5. 16. lis 2019.

    Come see our work on medical record notes modeling with Jonas Kemp and also on learning adaptive learning rates with Zhen Xu and at Baylearn today!

    1 proslijeđeni tweet 18 korisnika označava da im se sviđa
    Poništi
  6. 11. ruj 2019.

    Come work with us! Google is now accepting applications for the 2020 AI Residency Program, Healthcare! Head to for more details about the program. Applications close on Sept 17th, 2019! Questions? Go to

    14 proslijeđenih tweetova 42 korisnika označavaju da im se sviđa
    Poništi
  7. proslijedio/la je Tweet
    17. lip 2019.

    I'm very excited about our work on the use of machine learning and technology for healthcare at , & that we've hired David Feinberg () to lead our efforts in this space. Learn more about David and what we're doing in this interview.

    “Passion means we have to get this right, and feel that health is a cause worth fighting for, every day.” Meet David Feinberg, who recently joined Google to head up health efforts →
    50 proslijeđenih tweetova 256 korisnika označava da im se sviđa
    Poništi
  8. 15. lip 2019.

    See our work on learning graphical structure in medical records: Learning Graphical Structure of Electronic Health Records with Transformer for Predictive Healthcare at the Learning and Reasoning with Graph-Structured Representations workshop 3:30-4:30pm.

    1 reply 4 proslijeđena tweeta 9 korisnika označava da im se sviđa
    Poništi
  9. 14. lip 2019.

    Check out our work on medical records at the ICML workshops today! Analyzing the Role of Model Uncertainty for Electronic Health Records at the Uncertainty workshop and Time series modelling by restricting feature interaction at the Time series workshop.

    12 proslijeđenih tweetova 75 korisnika označava da im se sviđa
    Poništi
  10. 11. lip 2019.

    Come to the Google ICML booth to hear about some of our recent medical records research.

    Stop by the Google booth at 10:30AM today to chat with Research Engineer Charles Weill () about fast and flexible AutoML with AdaNet (), and Software Engineer Andrew Dai () about some of our health research efforts.
    Prikaži ovu nit
    2 proslijeđena tweeta 22 korisnika označavaju da im se sviđa
    Poništi
  11. 9. svi 2019.

    Come see Anna Huang give a talk on our music generation with transformer work at the poster session this morning!

    1 reply 9 proslijeđenih tweetova 52 korisnika označavaju da im se sviđa
    Poništi
  12. 8. svi 2019.

    Almost as many hands went up at for 'not everything can be learnt' vs. 'everything can be learnt'. Surprising for this audience!

    3 korisnika označavaju da im se sviđa
    Poništi
  13. proslijedio/la je Tweet
    7. svi 2019.

    TPU Pods are the hardware we use at for much of our research and production ML models for things like BERT, large-scale image classification, etc. They are now in beta on . Now you can get your own AI supercomputer by the hour!

    From 24 hours on-prem to 15 minutes on , Cloud TPU Pods accelerate ’s deep learning research while keeping operational costs low →
    51 proslijeđeni tweet 170 korisnika označava da im se sviđa
    Poništi
  14. proslijedio/la je Tweet
    7. svi 2019.

    At 4:15 at the Google booth, will talk about some of Google's health-related research efforts, including fairness in for health equity, scalable and accurate with electronic health records and more. We hope you'll join us!

    1 reply 18 proslijeđenih tweetova 95 korisnika označava da im se sviđa
    Poništi
  15. 20. velj 2019.

    In fact we showed that you don't even need additional data to get a gain from pretraining.

    First, the idea of using pre-trained language models for transfer learning was proposed in our paper in 2015: . (2/3)
    Prikaži ovu nit
    1 proslijeđeni tweet 16 korisnika označava da im se sviđa
    Poništi
  16. proslijedio/la je Tweet
    18. velj 2019.

    Since our work on "Semi-supervised sequence learning", ELMo, BERT and others have shown changes in the algorithm give big accuracy gains. But now given these nice results with a vanilla language model, it's possible that a big factor for gains can come from scale. Exciting!

    0:37
    We've trained an unsupervised language model that can generate coherent paragraphs and perform rudimentary reading comprehension, machine translation, question answering, and summarization — all without task-specific training:
    19 proslijeđenih tweetova 129 korisnika označava da im se sviđa
    Poništi
  17. proslijedio/la je Tweet
    19. velj 2019.
    Odgovor korisniku/ci

    Nice. But the idea of using pretrained language models AFAIK was first proposed by this paper (Also mentioned in Jacob's slides for the history around BERT: )

    1 reply 4 proslijeđena tweeta 26 korisnika označava da im se sviđa
    Poništi
  18. proslijedio/la je Tweet
    4. pro 2018.

    Great advice from Olivier Bousquet about being bold and creative in research. From test of time award talk at

    82 proslijeđena tweeta 309 korisnika označava da im se sviđa
    Poništi
  19. 12. lis 2018.

    If only we had transformer models back then to do bigger LM pretraining!

    I also want to point out that the masked LM idea is similar in principle to the idea of sequence autoencoders with input word dropout and embedding dropout and also LM that & I tried in 2015:
    Prikaži ovu nit
    13 korisnika označava da im se sviđa
    Poništi
  20. 22. kol 2018.

    The product of our Brain/Calico collaboration is now out!

    1/5 Recently finished a project applying deep learning to the peptide-spectra matching problem in proteomics . A great collaboration with and co. Overall, a ~10% improvement over SOTA.
    Prikaži ovu nit
    3 korisnika označavaju da im se sviđa
    Poništi

@iamandrewdai još nije objavio nijedan Tweet.

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:

    ·