Seminars and Events
Advances in Vision-and-Language Research Beyond Performance
Event Details
Vision-and-language models (e.g., CLIP) have achieved remarkable performance on a wide array of tasks. Nevertheless, with great power comes with great responsibility. In this talk, I will discuss two emerging research directions for vision-and-language models beyond performance—fairness and privacy. First, we identify a unique gender bias problem in image search and propose two simple yet effective solutions to debias the models during both training and inference. Then, we view language as a recipient to study multimodal model’s equal treatment of languages and investigate the multilingual accuracy disparity across other fairness dimensions including gender, race, and age. In the end, I will briefly talk about our recent efforts on building a privacy-preserving embodied agent to protect people’s privacy while serving them effectively.
Speaker Bio
Host: Muhao Chen, POC: Amy Feng
YOU ONLY NEED TO REGISTER ONCE TO ATTEND THE ENTIRE SERIES - We will send you email announcements with details of the upcoming speakers.
Register in advance for this webinar: https://usc.zoom.us/webinar/register/WN__0VhakI6Q6i3JsasdmNWcA.
After registering, you will receive an email confirmation containing information about joining the Zoom webinar.
Per the guest speaker, the recording for this AI Seminar talk will be posted on our USC/ISI YouTube page in about 4 months: https://www.youtube.com/user/USCISI.