Speaker Jane Wu - Apr 9 11am

Attend:

Wed Apr 9

• 11am - E2-180 - Special guest speaker - Jane Wu 
  • https://calendar.ucsc.edu/event/cse-colloquium-physical-reasoning-and-interaction-in-a-human-centered-world
  • This is an extra credit assignment in Canvas worth 1% of total class grade

Submit:

There are two parts to this assignment, each worth half the credit. Submit both in a single PDF. 

• 1 pt - Include a selfie of yourself clearly in the room that the presentation is happening in. Before or after the talk, just grab a snap on your phone with the screen in the background. Don't capture in the middle of the talk. 
• 1 pt - Answer the following using at least 1/2 page text total:
  • What were the main concepts presented in this talk?
  • What similar technologies have you heard about, or how do you think this relates to or could be used together with other technologies?
  • What new ideas does this bring up for you? Pretend you are an inventor, what are the next steps?

FAQ:

• But what if I have class at that time? 
  • Bummer. I can make learning opportunities available, but I can't make it work out for everyone everytime. Thats why its EC instead of required.
• Can I watch on Zoom?
  • Sure. But you will only be able to get half the EC for the write up. You have to show up in person to get the other half.
• What is this about again, am I interested?
  • Talks normally have a short Abstract and Bio provided by the speaker, to help you decide (I put it below). You can also google the speaker and check their web page to see what they work on.
  • Wednesday, April 9, 2025
    11:00 AM PT
    Engineering 2, 180
    UC Santa Cruz
    
    Physical Reasoning and Interaction in a Human-Centered World
    
    Dr. Jane Wu
    President’s Postdoctoral Fellow
    Department of Electrical Engineering & Computer Science
    UC Berkeley
    
    Abstract: While large machine learning models trained on Internet-scale data have recently demonstrated impressive generalization capabilities, existing approaches still struggle to reason about the geometry and dynamics of the physical world. Recovering humans and their interactions with objects from real world data is particularly challenging due to limited 3D datasets and complex contact and deformation. In this talk, I will present two novel paradigms that leverage strong structural and learned priors for 3D reconstruction at scale. First, I will present a weakly-supervised method for clothed human reconstruction that renders geometry via Marching Tetrahedra and image rasterization in an end-to-end differentiable manner. Second, I will present a scalable approach to handheld object reconstruction that builds on recent breakthroughs in large language/vision models and 3D object datasets. Finally, I will discuss how 3D perception of humans and human-object interactions lends itself to learning physical properties and manipulation policies for applications in AR/VR and robotics.
    
    Bio: Jane is a postdoctoral fellow in EECS at UC Berkeley, advised by Jitendra Malik and funded by the NSF Mathematical Sciences Postdoctoral Research Fellowship and the UC President's Postdoctoral Fellowship. She received her Ph.D. in CS from Stanford University, advised by Ronald Fedkiw in the Stanford AI Lab. She is a recipient of WiGRAPH Rising Stars in Computer Graphics (2022) and MIT Rising Stars in EECS (2024).