Bethge Lab

@bethgelab

Perceiving Neural Networks

Tübingen, Germany
Vrijeme pridruživanja: srpanj 2017.

Tweetovi

Blokirali ste korisnika/cu @bethgelab

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

  1. Prikvačeni tweet
    20. tra 2018.

    We are happy to announce the ! It's designed to inspire progress towards more robust vision models & real-world attacks. Great to team up with , and & to have as a sponsor!

    Prikaži ovu nit
    Poništi
  2. 31. sij

    Work together with Evgenia Rusak, Lukas Schott, , Julian Bitterwolf, Oliver Bringmann, ,

    Prikaži ovu nit
    Poništi
  3. 31. sij

    Does your DNN have problems with common corruption robustness? You can get suprisingly far by just training on noise! In our new paper, we evaluate how simple learned i.i.d. noise can help to generalize to ImageNet-C. Blog post @

    Prikaži ovu nit
    Poništi
  4. 16. sij
    Poništi
  5. proslijedio/la je Tweet
    18. ruj 2019.
    Poništi
  6. proslijedio/la je Tweet
    18. ruj 2019.

    Want to see how we bring interpretability to CNNs to predict V1 responses via end-to-end learning divisive normalization? Meet me at my poster W-56 With

    Poništi
  7. proslijedio/la je Tweet
    14. ruj 2019.

    : come check out our poster "The Notorious Difficulty of Comparing Human and Machine Perception" TOMORROW (Sunday, 17:15 - 20:15)! .

    Poništi
  8. 14. ruj 2019.
    Poništi
  9. proslijedio/la je Tweet
    14. ruj 2019.

    My colleague @MaxGuenthner talking about modelling V1 with divisive normalization at

    Poništi
  10. proslijedio/la je Tweet
    5. ruj 2019.

    the European Laboratory for Learning & Intelligent Systems (ELLIS) has now announced the 11 programs :

    Poništi
  11. proslijedio/la je Tweet
    13. kol 2019.

    We will be presenting our study on the effectiveness of low frequency in generating adversarial perturbations at the Adversarial ML morning session!

    Poništi
  12. 6. kol 2019.

    Happy to announce that our Robust Detection Benchmarking Toolkit is now part of the mmdetection toolbox! Benchmark:

    Poništi
  13. 18. srp 2019.

    Can your object detector handle noise? What will your autonomous vehicle do when it's foggy? And more importantly: Is your D(rago)NN ready for Snow when Winter is Coming? Find out with our robust detection benchmark (Pascal, Coco & Cityscapes):

    Poništi
  14. proslijedio/la je Tweet
    7. srp 2019.

    We developed gradient-based versions (L0, L1, L2 & Linf) of our Boundary adversarial attack that (1) resist gradient-masking, (2) perform better & are more query-efficient than SOTA (e.g. PGD or C&W) and (3) require virtually no hyperparameter tuning:

    Prikaži ovu nit
    Poništi
  15. proslijedio/la je Tweet

    Gradient descent will take any shortcut available to map inputs to targets. Human perception works differently: it starts from a different input (embodied stream vs static images), it doesn't have a target for each input, and it isn't trained with SGD.

    Poništi
  16. 1. srp 2019.

    Article discussing our recent work investigating the properties of current state-of-the-art algorithms. Hope to see more in this line of research!

    Poništi
  17. proslijedio/la je Tweet
    21. lip 2019.

    😎And now, 2.0 is published! 3D Markerless pose estimation of user-defined points across any species. Full step-by-step guide, Notebooks, & more! 👇🥳co-1st: &

    Prikaži ovu nit
    Poništi
  18. proslijedio/la je Tweet
    18. lip 2019.

    Hey tweeps. This week I started a new position as a research scientist at Amazon! I will miss all my colleagues from the , where I've learned so much. Now, I'm excited to apply vision science at Amazon scale!

    Poništi
  19. proslijedio/la je Tweet
    29. svi 2019.

    We just updated our "One-Shot Instance Segmentation" paper and code …! Now featuring additional few-shot results, a more robust way to evaluate and a lot of other improvements!

    Poništi
  20. proslijedio/la je Tweet
    10. svi 2019.
    Poništi
  21. 7. svi 2019.

    Today is our big day . Checkout Roberts oral on shape vs texture at 4pm. We also have several posters on robustness, interpretability and neuroscience: AM posters: #34 and #49 PM posters: #65, #69 and #78 Looking forward to seeing you

    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:

    ·