[CS] Re: Xiao Zhang Candidacy Exam/Oct 24

[Xiao Zhang via cs]

Hi all,

The Zoom link is available for anyone who prefers to join remotely.
https://uchicago.zoom.us/j/6432681009?pwd=bUllY1JOVE9objEwUE5QMkIySjUrZz09
<https://www.google.com/url?q=https://uchicago.zoom.us/j/6432681009?pwd%3DbUllY1JOVE9objEwUE5QMkIySjUrZz09&sa=D&source=calendar&ust=1730216558879674&usg=AOvVaw3KFYCX1W9n0S_dqj4BTvie>

Best,
Xiao

On Mon, Oct 14, 2024 at 12:08 PM via cs <cs at mailman.cs.uchicago.edu> wrote:

> This is an announcement of Xiao Zhang's Candidacy Exam.
> ===============================================
> Candidate: Xiao Zhang
>
> Date: Thursday, October 24
>
> Time: 2:00 -3:00pm CT
>
> Location: JCL 223
>
> Title: Representation Learning from and for Generative Models
>
> Abstract: In this talk, I will present my research on attempting to
> connect self-supervised representation learning and generative modeling,
> two crucial concepts in modern computer vision. I'll demonstrate how
> generative models acquire strong visual representations and how improving
> representation learning can further enhance image generation quality.
> Real-world images have complex visual structures, and for generative models
> to recreate them, they need to encode these visual representations
> internally. We validate this by developing a scalable compression technique
> that extracts meaningful low-dimensional semantic representations from all
> layers of deep generative models. This also helps us interpret the internal
> workings of these models and reveals that their computational pathways
> resemble the 'what' and 'where' visual processing paths
> found in human perception. To further enhance representation learning in
> generative models, we identify a key design flaw in residual connections
> that hinders generative feature learning. We address this with a new
> network design, decayed residual connections, which gradually reduces the
> influence of skip connections in residual networks, promoting low-rank
> representations in the bottleneck. This design significantly boosts feature
> learning in masked autoencoders and improves the generation quality of
> diffusion models, all without adding new parameters.
>
> Advisors: Michael Maire
>
> Committee members:
> Michael Maire, Rebecca Willett, David Forsyth, Greg Shakhnarovich, Anand
> Bhattad
>
> When unsubscribing, use your cnetid at cs.uchicago.edu address if your
> cnetid at uchicago.edu does not work.
>
> cs mailing list  -  cs at mailman.cs.uchicago.edu
> Edit Options and/or Unsubscribe:
> https://mailman.cs.uchicago.edu/mailman/listinfo/cs
> More information here: https://howto.cs.uchicago.edu/techstaff:mailinglist
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.uchicago.edu/pipermail/cs/attachments/20241024/fd23273b/attachment.html>