Jeffrey De Fauw

@JeffreyDeFauw

Research in Deep Learning at , and work in health research, focussing on clinically applicable research in several domains. Opinions my own.

London, England
Vrijeme pridruživanja: lipanj 2015.

Tweetovi

Blokirali ste korisnika/cu @JeffreyDeFauw

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

  1. Prikvačeni tweet
    1. sij

    Very excited to share where we show an AI system that outperforms specialists at detecting breast cancer during screening in both the UK and US. Joint work with and published in today!

    Prikaži ovu nit
    Poništi
  2. proslijedio/la je Tweet
    26. sij

    Teaching Deep Unsupervised Learning (2nd edition) at this semester. You can follow along here: Instructor Team: , , , Wilson Yan, Alex Li, YouTube, PDF, and Google Slides for ease of re-use

    Poništi
  3. proslijedio/la je Tweet
    22. sij

    Be sure to try the interactive demos. E.g. You can click through individual neurons and see their structure, or enable them all to see how they connect together to form more complex structures:

    Prikaži ovu nit
    Poništi
  4. proslijedio/la je Tweet
    15. sij

    We have 2 papers published in today! 🎉 One describes AlphaFold, which uses deep neural networks to predict protein structures with high accuracy. AlphaFold made the most accurate predictions at the 2018 scientific community assessment CASP13. 1/4

    Prikaži ovu nit
    Poništi
  5. proslijedio/la je Tweet

    The fact it generalize in this way has promising implications for the eventual use of such systems in the real world.

    Poništi
  6. proslijedio/la je Tweet
    1. sij

    For insights on why the paper is a step forward, see 's excellent thread on tuples, delving deep, and all the complexities of this work:

    Prikaži ovu nit
    Poništi
  7. 1. sij
    Poništi
  8. 1. sij

    Some slightly less technical summary of our mammography work in our author notes:

    Poništi
  9. proslijedio/la je Tweet
    1. sij

    In today, a large UK/US retrospective study of deep learning for improving accuracy of mammography/breast diagnosis c/w radiologists by and collaborators by

    , , i još njih 5
    Prikaži ovu nit
    Poništi
  10. 1. sij

    Even given all these nuances, I’m still convinced that AI will have a strong positive impact on our lives, of which health will be an important aspect. I hope our work can contribute to that in the longer-term! 16/16

    Prikaži ovu nit
    Poništi
  11. 1. sij

    In the end, for AI health we need well-designed clinical trials to validate performance but this takes time. Note that these clinical trials _still_ have to think about a very similar (dataset, metric, baseline). 15/16

    Prikaži ovu nit
    Poništi
  12. 1. sij

    We also show a reader study on the US dataset -- performed by an external research organization, which is specialized to do clinical trials and evaluations. There is plenty more detail that is obfuscated by the technical language of a paper, I’ve tried to uncover some. 14/16

    Prikaži ovu nit
    Poništi
  13. 1. sij

    Note a subtlety: the radiologists are “gatekeepers” for further investigation. This means that if they don’t follow up, there is no possibility to confirm any malicious mass. We discuss this more in e.g. Extended Data Fig 4. 13/16

    Prikaži ovu nit
    Poništi
  14. 1. sij

    We can basically compare directly by putting the AI system in the exact same part of the clinical pathway: woman comes for screening, radiologist(s) decide if follow-up is needed vs AI decides if follow-up is needed. It’s similar for the US but we also provide another eval. 12/16

    Prikaži ovu nit
    Poništi
  15. 1. sij

    … it’s the result of countless hours of discussions with both ML and clinical collaborators. 3. Baseline performance: for the UK we have the _original_ decisions from the radiologists, together with the results of arbitration if relevant. 11/16

    Prikaži ovu nit
    Poništi
  16. 1. sij

    2. Metric: we look at biopsy proven cancer within 39 months (UK) and 27 months (US). We say it’s not cancer when we have at least one follow-up visit confirming this. These definitions are not trivial -- and there is a lot of nuance -- …. 10/16

    Prikaži ovu nit
    Poništi
  17. 1. sij

    The US dataset represents a different type of screening program (1-2 years screening, single reader). The supplementary data has plenty of tables and statistics to reveal the level of detail that is hidden away. 9/16

    Prikaži ovu nit
    Poništi
  18. 1. sij

    Concretely about this for our paper: 1. Datasets: the UK dataset (from ) is a representative sample directly from breast screening, we have spent a very large amount of time doing this thoroughly. 8/16

    Prikaži ovu nit
    Poništi
  19. 1. sij

    … dataset has unreasonable exclusion criteria, dataset is not randomly sampled from a clinical pathway, baseline performance is post-hoc, ground truth for metric is biased, etc. There are loads and, some can be very subtle. 7/16

    Prikaži ovu nit
    Poništi
  20. 1. sij

    Unfortunately that means to fully understand work in AI health, you need to delve quite deep. Examples of what invalidates the mapping for my interpretation (things I look out for): dataset is artificially constructed (“let’s find 100 examples of cancer”), … 6/16

    Prikaži ovu nit
    Poništi
  21. 1. sij

    Each of (dataset, metric, baseline) is critical in understanding the actual clinical impact of results. Seemingly small details can completely change the impact to the point where many feel the summary is invalid (will not translate in practice). 5/16

    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:

    ·