Maurice Weiler

@maurice_weiler

AI researcher with a focus on geometric and equivariant deep learning. PhD candidate under the supervision of Max Welling. Master's degree in Physics.

Vrijeme pridruživanja: siječanj 2018.

Tweetovi

Blokirali ste korisnika/cu @maurice_weiler

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

  1. Prikvačeni tweet
    11. lip 2019.

    Come to our talk on Gauge Equivariant Convolutional Networks and Icosahedral CNNs today at 14:40 @ Grand Ballroom, . Happy to discuss more details and connections to physics at poster #76 @ Pacific Ballroom, 18:30. With , and .

    Poništi
  2. proslijedio/la je Tweet
    26. sij

    Quaternions and Euler angles are discontinuous and difficult for neural networks to learn. They show 3D rotations have continuous representations in 5D and 6D, which are more suitable for learning. i.e. regress two vectors and apply Graham-Schmidt (GS).

    Prikaži ovu nit
    Poništi
  3. proslijedio/la je Tweet
    21. sij

    We are organizing an Workshop on Geometric and Relational Deep Learning! Registration invites will be shared soon. Interested in participating? Consider submitting an abstract or get in touch: w/

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

    For those images aren't all taken in a standard orientation (all of us!), this is a super interesting library. Their GIF shows how their feature fields are invariant with respect to rotation, unlike a standard CNN. your kind of paper

    Prikaži ovu nit
    Poništi
  5. 10. sij

    Our work on Gauge Equivariant CNNs made it into Quanta Magazine! The article gives a nice overview on coordinate independent convolutions and connections between theoretical physics and deep learning.

    Poništi
  6. proslijedio/la je Tweet
    22. pro 2019.

    Really exciting to have my first paper accepted at ! It provides the first group theoretical approach towards equivariant visual attention. Nice things coming up next! . Co-Attentive Equivariant Nets:

    Poništi
  7. proslijedio/la je Tweet
    21. pro 2019.

    Really happy with my paper being accepted ! It describes a flexible framework for building G-CNNs that are equivariant to a large class of transformation groups. B-Spline CNNs on Lie groups:

    Poništi
  8. 21. pro 2019.

    Check also 's work on scale equivariance. It makes the non-invertibility of dilations on pixel grids explicit via scale space theory and semi-groups.

    Prikaži ovu nit
    Poništi
  9. 21. pro 2019.

    The formulation is very similar to our rotation equivariant Steerable Filter CNNs: both models use group convolutions and define kernels in the continuum to group-transform them exactly via steering.

    Prikaži ovu nit
    Poništi
  10. 21. pro 2019.

    Nonetheless the proposed models work very well. It seems like the gains due to equivariance are outbalancing the loss of energy from kernels running out of the scale range.

    Prikaži ovu nit
    Poništi
  11. 21. pro 2019.

    A nice work on scale-equivariant CNNs via group convolutions. In contrast to e.g. SO(2), the dilation group (R^+,*) is non-compact. A conv over scales thus needs to be restricted to a certain range which introduces boundary effects similar to the zero padding artifacts of CNNs.

    Prikaži ovu nit
    Poništi
  12. 11. pro 2019.

    1st line: E(2)-equivariant CNN 2nd line: conventional CNN paper: code: docs:

    Prikaži ovu nit
    Poništi
  13. 11. pro 2019.

    Check out our poster #143 on general E(2)-Steerable CNNs tomorrow, Thu 10:45AM. Our work solves for the most general isometry-equivariant convolutional mappings and implements a wide range of related work in a unified framework. With

    Prikaži ovu nit
    Poništi
  14. proslijedio/la je Tweet
    25. lis 2019.

    Check out "Scale-Equivariant Steerable Networks" (). It is joint work with Michał Szmaja and Arnold Smeulders. We build scale-equivariant CNNs which do not use image rescaling and do not limit the admissible scale factors.

    Prikaži ovu nit
    Poništi
  15. proslijedio/la je Tweet
    24. ruj 2019.

    Since I keep getting questions about the cotan formula for tet meshes, I wrote up a note about the n-dimensional cotan formula (including a nice expression for tet meshes). Hopefully this saves me infinitely many emails for n ≥ 4.

    Poništi
  16. proslijedio/la je Tweet
    18. ruj 2019.

    “You get away from coordinates, which is like the lowest level of number, and elevate the idea of number to incorporate geometry.” —Charlie Gunn on using geometric algebra for computer vision, graphics, and engineering. See

    Poništi
  17. proslijedio/la je Tweet
    27. kol 2019.

    Please consider applying for this tenure track position on fairness, accountability and transparency in AI.

    Poništi
  18. proslijedio/la je Tweet
    23. kol 2019.

    WFC running at interactive speed on that funky grid. Next up is to make it deterministic

    Prikaži ovu nit
    Poništi
  19. proslijedio/la je Tweet
    13. kol 2019.

    Gauge Equivariant Convolutional Networks and the Icosahedral CNN

    Poništi
  20. proslijedio/la je Tweet
    28. lip 2019.

    Very excited to share our work 'GEAR: Geometry-Aware Rényi Information' (w/ @nevitus, M.Schwarzer & ): TL;DR: A perspective on Information Theory which takes geometry into account with apps. in image barycenters, mode counting & gen. models.

    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:

    ·