I’m attending NeurIPS 2024, and within five seconds of walking in the door, there’s a happy lab reunion! Left is Rob Brekelmans. After finishing his PhD with me and Aram at USC, he took a prestigious postdoc fellowship at Vector Institute. After his 2024 ICML best paper award, you better believe he’ll get snapped up on the faculty hiring marking, so better move quickly if you’re hiring! On the right side is Sami Abu-el-Haija, a wizard with graph neural networks, who is now working at Google Research.

Let’s get back to the reason I’m here. Here’s a nice video from the lead author Yunshu Wu:  (though I think to see the video you may have to be registered for the conference). Unfortunately she couldn’t make it to the conference so I’m presenting in her place. It’s a fun and innovative paper making connections between diffusion models, log likelihood ratios, and (noise) classification.

The line of work being exploited here was developed in 2 papers (ICLR 2023, ICLR 2024) by Xianghao Kong. He just successfully defended his thesis proposal, so will also be on the market in June. He’s already done some great industrial research on diffusion models too, at Sony and Adobe, so I’m sure he’ll be in high demand.

Finally, to round out the lab news, I’d like to point out a really cool preprint by a new student, Shaorong Zhang, on diffusion bridge models.  And I’m delighted to announce that two more of my USC students: Neal Lawton and Kyle Reing, successfully defended their theses. Congrats!

For Google search, I finally got a new professional portrait, thanks to the UCR photographer Stan Lim. He also does the photography for the prison BA program that my wife helped to start, UCR Lifted, so if you are interested in prison education, you must check out the cool shots there.

Greg Ver Steeg
Associate Professor
CSE
(UCR/Stan Lim)