Visitor's Talk



Escher Pattern Through Hourglasses - Temporal Analysis and Pattern Discovery in Multimedia Sequences

• Speaker: Dr. Lexing Xie, IBM Research

• Abstract:
  We are witnessing a proliferation of large multimedia collections with the recent advances in hardware and networking infrastructures. The sheer volume of these content exceed the human capability to annotate and filter them manually, yet our ability to automatically process, interpret, and use these rich corpora has notably lagged behind. As a result parsing and recognizing objects and concepts from images has recently received much attention in computer vision and multimedia. Our investigations are on two aspects that compliments supervised visual recognition: (1) unsupervised pattern discovery, i.e., finding recurrent segments from multimedia sequences with a small amount of domain knowledge and little annotation; (2) temporal modeling in video for mining and recognition in different scales. In the first aspect, we design dynamic graphical models to capture patterns in video, and we explain the discovered patterns via statistical association to multiple modalities. In the second aspect we present models for learning inter-shot temporal dependence as well as intra-shot temporal variations. We will present evaluations on broadcast news and sports programs, and discuss its extension to long-term temporal modeling and mining, and to a broader context of mining in multimedia and other data types.

• Biographical Sketch:
  Lexing Xie is a researcher in IBM T. J. Watson Research Center in Hawthorne, NY. She is interested in making multimedia content more accessible and useful to people, as well as understanding people to help build better models and systems that can process multimedia repositories. Her research has been on multimedia signal processing, content analysis, pattern mining, machine perception and learning. She received PhD in Electrical Engineering from Columbia University in 2005, and B.S. in EE from Tsinghua University, Beijing, China in 2000. She has worked in Microsoft Research Cambridge June-August 2003. She is the recipient of the 2005 IBM Research Josef Raviv Memorial Postdoc fellowship, she has also received best student paper awards from IEEE ICIP in 2004 and ACM Multimedia in 2002 and 2005, respectively.