Andreas Terzis

@aterzis

Googler, X-Prof .jhu aterzis@google.com

Mountain View, CA
research.google/people/Andreas…
Joined February 2008
17 Photos and videos Photos and videos

Tweets

You blocked @aterzis

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

  1. Jan 10

    Want to get an intuitive understanding about how ML models trained with differential privacy can protect the training data? This explorable will help! The post also describes many of the challenges in applying differential privacy to ML models.

    Training with differential privacy stops models from inadvertently leaking sensitive data, but there’s an unexpected side-effect: reduced accuracy on underrepresented subgroups.
    1 like
    Undo
  2. Retweeted
    Jan 1

    Discovered this Georgian mineral water. It tastes like sparkling pool water, but somehow good?? Contents/Liter: Bicarbonate 4000mg Sodium 1500mg Calcium 50mg Magnesium 50mg *Chlorine* 350mg

    10 replies 1 retweet 24 likes
    Undo
  3. 31 Dec 2021

    Covariance-Aware Private Mean Estimation Without Private Covariance Estimation by (4/4)

    3 likes
    Show this thread
    Show this thread
    Undo
  4. 31 Dec 2021

    Differential Privacy and the 2020 Census in the United States by (3/4)

    1 reply 5 likes
    Show this thread
    Show this thread
    Undo
  5. 31 Dec 2021

    Private Histograms in the Shuffle Model by Albert Cheu: (2/4)

    1 reply 2 likes
    Show this thread
    Show this thread
    Undo
  6. 31 Dec 2021

    The last three talks of the year from our Privacy in ML seminar organized by (1/4)

    1 reply 3 retweets 19 likes
    Show this thread
    Show this thread
    Undo
  7. 24 Dec 2021

    How to deploy machine learning with differential privacy by Abhradeep Thakurta and

    8 likes
    Undo
  8. 23 Dec 2021

    Sad day

    I would just pause to note that this is an epic photograph, potent with symbolism, that is going to go down in history. Taken by Tyrone Siu of Reuters ⁦⁩
    Show this thread
    2 likes
    Undo
  9. Retweeted
    9 Nov 2021

    Happening tomorrow at 9am US pacific time!! Please RSVP and tune in if you are interested in learning about federated learning, federated analytics, differential privacy, model personalization, and TensorFlow Fedeated. I promise this event will be fun and rewarding 🤓😎💥

    Do you want to master Federated Learning and Analytics, Personalization, Differential Privacy, and Federated? We are hosting a one-of-a-kind workshop on Nov 10 with thrilling talks and discussions. Register now to reserve a spot!
    3 retweets 23 likes
    Undo
  10. Retweeted
    2 Nov 2021

    Do you want to master Federated Learning and Analytics, Personalization, Differential Privacy, and Federated? We are hosting a one-of-a-kind workshop on Nov 10 with thrilling talks and discussions. Register now to reserve a spot!

    , , and 4 others
    37 retweets 131 likes
    Undo
  11. Retweeted
    26 Oct 2021

    In a new blog post with my colleague Abhradeep Thakurta, we survey the practice of designing private machine learning algorithms via differential privacy. Some highlights / takeaways I found important:

    2 replies 30 retweets 215 likes
    Show this thread
    Show this thread
    Undo
  12. 25 Oct 2021

    An accessible yet comprehensive technical introduction into private ML algorithms through differential privacy

    New post about practical machine learning with differential privacy by my colleagues and Abhradeep Thakurta.
    3 retweets 10 likes
    Undo
  13. 14 Oct 2021

    See also ' talk on the same topic:

    a screenshot of the full title, authors and abstract of https://arxiv.org/abs/2110.06452
    Hey this looks great 👀
    4 likes
    Undo
  14. Retweeted
    8 Oct 2021

    In joint work with , we propose an approach for tuning hyperparameters with Renyi Differential Privacy! Most results on deep learning with DP report the privacy budget of the single best performing training run, rather than the budget expended by the entire search

    1 reply 21 retweets 102 likes
    Show this thread
    Show this thread
    Undo
  15. Retweeted
    7 Oct 2021

    "Fast Sample Reweighting" is a new paper from our research group that allows you to re-weight training samples effectively without the need for additional unbiased reward data. PS: We’re hiring!     

    1 reply 8 retweets 55 likes
    Undo
  16. Retweeted
    2 Oct 2021

    Thanks for sending author copies! If you are a student anywhere in the world that would like a free copy, reply to this tweet. I will pick 2 lucky winners by the end of next week. RT to share!

    200 replies 101 retweets 411 likes
    Show this thread
    Show this thread
    Undo
  17. 30 Sep 2021

    Thank you to for running the seminar and to the invited speakers for the great talks!

    We just uploaded 6 more videos from our Privacy in ML seminar at Google Research. In no particular order: 1/6 Gautam Kamath CoinPress: Practical Private Mean and Covariance Estimation
    Show this thread
    4 likes
    Undo
  18. Retweeted
    30 Sep 2021

    The Art of Linear Algebra [pdf]

    2 replies 48 retweets 181 likes
    Undo
  19. Retweeted
    29 Sep 2021

    How can we productively work toward creating safe machine learning models? After struggling with this question for the past several years, we have developed a new roadmap for ML safety. Post: Paper:

    In the full paper we also consider the problem of "External Safety" to mitigate risks to how ML systems are handled.
    Robustness.
    Monitoring.
    Alignment.
    , , and 2 others
    2 replies 56 retweets 235 likes
    Undo
  20. 10 Sep 2021

    Story about poop detection for Roomba: Chat in our group: "What about adversarial poop?" That's my crew :-)

    1 reply 4 retweets 21 likes
    Undo

@aterzis 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

    ·