Mathias Unberath

@MathiasUnberath

Assistant Professor working on Imaging, Machine Learning, and Augmented Reality to develop collaborative systems for healthcare.

Baltimore, MD
Joined January 2015
26 Photos and videos Photos and videos

Tweets

You blocked @MathiasUnberath

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

  1. Retweeted
    Jan 2

    and communities need to interact more to productively bring about intelligent systems that intend to assist people in real-world settings. A systematic review of the literature on transparent ML for medical image analysis reveals a clear disconnect btw the communities

    Transparent ML for MIC zeros in on method development & forgets human factors that determine intelligibility and downstream goals, e.g., trust. We suggest prioritizing formative user research to ensure that ML systems afford transparency.
    Show this thread
    2 retweets 8 likes
    Undo
  2. 25 Dec 2021

    Increasing the acceptance of empirical formative user research as an integral component of human-centered ML design for MIC, will be critical in ensuring that the assumptions on which human-centered systems are built hold in the real world. Our guidelines attempt a first step.

    1 reply 3 likes
    Show this thread
    Show this thread
    Undo
  3. 25 Dec 2021

    In short, we find that contemporary research on transparent ML for medical image analysis at risk of being incomprehensible to users, and thus, clinically irrelevant.

    1 reply 4 likes
    Show this thread
    Show this thread
    Undo
  4. 25 Dec 2021

    Further, while all studies considered medical image analysis, only ~half of them were developed by multidisciplinary clinician-engineering teams. And only 3 (of the 68) reported empirical user testing.

    1 reply 1 like
    Show this thread
    Show this thread
    Undo
  5. 25 Dec 2021

    From review, we found that only ~half of the studies specified who users will be. Those that do specify target clinicians (no other stakeholders considered). However, no study considered formative user research to understand user context, needs, wants, and priors.

    1 reply 1 like
    Show this thread
    Show this thread
    Undo
  6. 25 Dec 2021

    Transparency is not a property of an ML model, but an affordance (relationship between model and user). In healthcare, there is a large knowledge gap between devs and potential targets (clinicians, patients,...). So, transparent - to whom?

    1 reply 1 like
    Show this thread
    Show this thread
    Undo
  7. 25 Dec 2021

    Transparent ML for MIC zeros in on method development & forgets human factors that determine intelligibility and downstream goals, e.g., trust. We suggest prioritizing formative user research to ensure that ML systems afford transparency.

    INTRPRT: A Systematic Review of and Guidelines for Designing and Validating Transparent AI in Medical Image Analysis by Haomin Chen et al. including
    1 reply 8 likes
    Show this thread
    Show this thread
    Undo
  8. Retweeted
    14 Dec 2021

    Engineers at have recently developed a new -based system that could be used to train surgeons to complete skull base surgeries, as well as potentially other complex surgical procedures.

    6 retweets 9 likes
    Undo
  9. 3 Dec 2021

    We systematically review non-adversarial robustness and suggest a struct causal model of imaging. Robustness then measures how well model performs on counterfactually altered images, where rare phenomena are artificially emphasized via soft interventions.

    3 retweets 11 likes
    Undo
  10. Retweeted
    16 Nov 2021

    We are hiring! is looking for TT faculty in natural language processing (including machine reasoning and grounded language), machine learning, AI and HCI. Join us!

    9 retweets 8 likes
    Undo
  11. Retweeted
    27 Oct 2021

    Pediatric Otoscopy Video Screening with Shift Contrastive Anomaly Detection by Weiyao Wang et al. including

    3 retweets 8 likes
    Undo
  12. Retweeted
    12 Oct 2021

    Check out Stereo Transformer at , which revisists the stereo depth estimation from a sequential problem. We will be there today and Thursday for Q&A. Many thanks to my collegues/advisor and more!

    4 retweets 13 likes
    Show this thread
    Show this thread
    Undo
  13. Retweeted
    29 Sep 2021

    The video for our paper is now also available on the youtube!

    She surprised me with the idea and I loved it A true testament to Anna creativity in communication and research. BTW, she's on the market for a PhD position last I heard :)
    1 retweet 7 likes
    Undo
  14. 29 Sep 2021

    She surprised me with the idea and I loved it A true testament to Anna creativity in communication and research. BTW, she's on the market for a PhD position last I heard :)

    Of all the 2021 paper videos I have seen so far, nothing comes close to 's in terms of sheer creativity. A pure joy to watch and highly recommended. P.S. I loved the NN eating CTs.
    8 likes
    Undo
  15. 27 Sep 2021

    Delightful start to with Adnan and winning an Outstanding Paper Award in the workshop for our work on immersive VR + haptics to train experts and generate data for ML at the same time.

    1 retweet 26 likes
    Undo
  16. Retweeted
    22 Sep 2021

    We have prepared super cool video for the upcoming - drop by us at the poster to chat about the interpretability of the severity scoring in pelvic traumas 📝:

    2 replies 1 retweet 16 likes
    Undo
  17. Retweeted
    14 Sep 2021

    On the Sins of Image Synthesis Loss for Self-supervised Depth Estimation by et al. including

    5 retweets 6 likes
    Undo
  18. Retweeted
    16 Aug 2021

    An Interpretable Algorithm for Uveal Melanoma Subtyping from Whole Slide Cytology Images by Haomin Chen et al. including

    2 retweets 8 likes
    Undo
  19. Retweeted
    8 Aug 2021

    Come and work with us on robot-assisted surgery in oncology, PhD position available, apply now:

    9 retweets 20 likes
    Undo
  20. 6 Aug 2021

    2D/3D registration is at the center of image-based surgical navigation and will enable mixed reality and robotic techniques. But, it's notoriously difficult. 😢 We review how machine learning helps and identify next frontiers.

    1 reply 3 retweets 21 likes
    Undo

@MathiasUnberath 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

    ·