Powerful Generative Models and Responsible AI: Advancing Video Understanding for Real-World Applications
COM1 Level 2
Cerebro@SoC, COM1-02-05
Abstract:
Powerful generative models have revolutionized the field of AI and ignited the business world, but their potential for misuse has also raised global concerns. Responsible AI, which has been gaining attention thanks to real-world productization of AI, has become increasingly critical in the safe deployment of these models. In this public forum, I will share insights and considerations from a research perspective, focusing on the remaining challenges that academia could lead in investigating. I will provide an in-depth overview of our research in video understanding, covering trimmed to untrimmed video analysis, offline to online setting, few-shot to weakly supervised learning, and unimodal to multimodal analysis. Our work not only achieves SOTA performance on benchmarks across all settings, but also integrates seamlessly to support complex real world application scenarios.
Bio:
Mei Chen is a Principal Research Manager at Microsoft Cloud & AI. She leads ROAR (Responsible & Open Ai Research) to build state-of-the-art multimodal multilingual technology that is context-aware and culturally sensitive to serve diverse customer needs. ROAR's models have been shipped in new Bing powered by LLM, Bing Image Creator powered by DALL-E 2, Office Designer powered by DALL-E 2, Azure OpenAI Service , Github Copilot, and the Azure AI Content Safety Service . ROAR’s research has consistently been published at top conferences CVPR, ICCV, and ECCV. From 2018 to 2020, Mei served as Chief Scientist to the Corporate Vice President for Computer Vision at Microsoft. From 2014 to 2018, she was an Associate Professor at the State University of New York at Albany. From 2011 to 2014, she built and led the Intel Science & Technology Center on Embedded Computing (ISTC-EC) hosted at Carnegie Mellon with $7.5 Million in research funding. Mei’s work in computer vision and biomedical image analysis won three Best Paper Awards. She holds over 30 patents from the USPTO. Her services to the research community include journal editorial for IJCV and MVA, Conference Chair at CVPR and ICVS, Area Chair at CVPR, ICCV, and Workshop Chair for Multimodal Content Moderation and Computer Vision for Microscopy Image Analysis at CVPR. She earned a Ph.D. in Robotics from the School of Computer Science, Carnegie Mellon University, and an M.S. and B.S. from Tsinghua University, Beijing, China.