Cagri Dagli, a current ECE graduate student; ECE alumni Nemanja Petrovic and Shyamsundar Rajaram; and ECE Professor Thomas S. Huang, were recently recognized at the 2007 IEEE International Workshop on Semantic Learning Applications in Multimedia (SLAM) with a Best Paper Award. The SLAM workshop took place during the Computer Vision and Pattern Recognition (CVPR) conference in Minneapolis on June 18. Their paper entitled, “Diverse Active Ranking for Multimedia Search” was similarly honored at the International Conference on Pattern Recognition in Hong Kong last summer.
“Our work is essentially about trying to implicitly model user behavior in an information-retrieval scenario,” said Dagli. “Imagine going to Google images to search for a particular concept and being able to give feedback on the search results.”
Huang, the William L. Everitt Distinguished Endowed Professor in Electrical Engineering, explained that several years ago the group pioneered the use of ‘relevance feedback’. “It is an iterative procedure of learning, where at each iteration the computer will present several images to the user, and ask him or her to label each as either relevant or irrelevant,” said Huang. “Through this procedure, the computer tries to quantify the user's idea of ‘relevance’. Our SLAM paper presents a technique which combines ‘active learning’ with relevance feedback. By our algorithm, the computer knows which set of images it should present to the user to get maximum information — with regard to the user's idea of ‘relevance’ — from the feedback.”
The group was proud to have received the award. “It's flattering to be recognized for our work,” said Dagli. “This particular paper evolved organically from the highly free and collaborative nature of Professor Huang's research group.”
The Best Paper Award was sponsored by the IEEE Pattern Analysis and Machine Intelligence Technical Committee. It is awarded in conjunction with CVPR conference. Nearly 1,500 people attended CVPR and its thirteen associated workshops. Each individual workshop had from 50 to 100 participants. For the SLAM Workshop, 25 papers were submitted, of which 12 were accepted. All submitted papers were reviewed at least twice by the thirty members of the program committee, and selection was based on “relevance to the workshop, novelty, and technical quality.” The audience then voted for the best paper.
According to the SLAM Web site, the workshop strives to “bring together an interdisciplinary group of researchers in computer vision, speech/music recognition, knowledge representation and ontologies, machine learning, natural language, and other areas to examine the issues and recent results in using semantic knowledge to enhance multimedia.”
Rajaram, a 2007 PhD graduate, now works for HP Labs. Petrovic completed his PhD in 2004 and now works at Google.