SenseTime Talk on Computer Vision
Thursday, December 14, 2017 at 11:00am to 1:00pm
32-G449 - MIT Stata Center
Please join SenseTime, the company Forbes called "the unicorn of AI," for a talk led by Dr. Xiaoou Tang PhD '96 (XIII W) and Dr. Dahua Lin PhD '12 (VI D). They will discuss how computer vision and AI-powered deep learning help us recognize faces and objects, and understand the world.
A light lunch will be served.
AI and Computer Vision: From Academia to Industry
Based on the research experience at the Chinese University of Hong Kong and the industrial development at Sensetime, a leading AI company in China, Professor Xiaoou Tang will discuss the advantage of combining the best of both worlds, academia research and industrial application. He will also talk about how Hong Kong becomes such a unique place for AI research and development. He will discuss the making of SenseTime.
From Images to Text: A Journey Beyond Performance
I will introduce our efforts on image captioning, a topic that receives increasing attention in recent years. Conventional approaches to this task often encourage reproduction of training patterns (a safe way to obtain high scores in classical metrics), while overlooking other important qualities of human languages, e.g. naturalness, coherence, diversity, and distinctiveness. In our recent work, we study the limitations of the mainstream training strategies and evaluation policies, and propose several alternative methods to captioning. We show that these methods can effectively improve the generated captions from different aspects. At the end of the talk, I will also share my reflections on several prevalent guidelines in computer vision, e.g. deep architectures and end-to-end learning.