The kind of intense team contests found in sporting events on campus aren’t usually played out in academic competitions. But the Grand Finals of Star Challenge, an international multimedia search competition worth $100,000 to the winning team, took on a decidedly competitive tone. And Beckman Institute researcher Tom Huang sounds just like a coach when talking about his team’s effort.
The University of Illinois team, headed by Huang and fellow Beckman faculty member Mark Hasegawa-Johnson, was the only entry from the United States to reach the Grand Finals held in Singapore in October and was ahead of the other four finalists going into the last round of competition. Prior to the trip to Singapore, Huang sounded like a coach who was just happy to reach the finals. After the Illinois team took home the Bronze medal, it was clear they had their sights set on first place.
“The one thing I want to emphasize is that we entered this kind of competition mainly to learn, not really to win,” Huang said prior to the finals. “Winning means having some luck, especially this final competition, which really depends on luck.”
Afterwards, the spirit of competition was evident.
“We are happy and I’m really proud of the team but of course we were disappointed,” Huang said.
The sort of team competition found in Star Challenge is somewhat unusual in academia. Huang said he and his students participate in other evaluations but Star Challenge was different because of the teamwork it required and because of its competitive nature – the Grand Finals was a two-hour, on-stage event with music playing over loudspeakers and a crowd of interested onlookers.
“My students work together but not to this extent,” Huang said. “It was a very intense experience.
“There was an audience and an emcee who was very dramatic and trying to make it exciting,” he added with a laugh. “With all the music in the background it was a little distracting.”
Star Challenge was created as part of the opening of Fusionopolis, the new research and development arm of Singapore’s Agency for Science, Technology and Research (A*STAR). There were 56 teams from around the world originally entered in the competition, which over a nearly 10-month process required entrants to create algorithms that could search Singapore television shows for specific audio and video segments.
The Illinois team went by the name UIUC-YX and, in addition to Huang and Hasegawa-Johnson, included students Jui Ting Huang, Dennis Lin, Xi Zhou, Zhen Li, Xiaodan Zhuang, and team leader Yuxiao Hu. Although they were leading after the qualifying rounds, the Illinois team was thrown a curve in the last round: Second Life, the user-generated virtual world, was going to be a part of the Grand Finals. At a planning meeting for the finals held a few days before they left, members of the team were asked how many had used Second Life. Muted laughter was the only response.
“It was a complete surprise,” Huang said of the Second Life angle. “We have just started to look into it.”
But as it turned out, the biggest obstacle in the Grand Finals was the mix of languages found on Singapore television. The mission in the finals was to find specific sounds such as phonemes or words in the audio track of the videos.
“You want to find segments where the audio contains these phonemes but it could be spoken by anybody,” Huang said. “The second query was a segment of speech. You want to find video segments containing these words but spoken by different people. The query was usually very short, one word or two words, so it was difficult to find matches.”
Add in the fact that members of the team didn’t speak some of the languages and that made the task even harder. Despite that, the Illinois team led right up until the end. Students took turns with the tasks during the two-hour competition, with three team members on stage at a time.
“I did not go up on stage but Mark did,” Huang said. “We changed people all the time, depending on the task. They give the score along the way and 15 minute before the end we were leading. We stayed No. 1 until about five minutes before the end.”
Other teams contributed answers near the end, with a team from the National University of Singapore taking the top prize. Despite the heated competition, Huang said his team members weren’t nervous, unlike his wife Margaret, who also made the trip.
“They were too busy to be nervous,” he said. “Actually my wife was the most nervous person. I don’t know why but she was really, really intense.”
Huang said that a major difference between the Grand Finals and the first three rounds was that in the earlier rounds all of the search methods were totally automatic (no human intervention) while in the Finals, the contestants were allowed to listen to the audio and look at the video, giving an advantage to teams that were familiar with the languages and the TV program contents or formats.
Even though the competition was exciting, the purpose behind Star Challenge was scientific, both for the sponsors and the participants. The competition was aimed at advancing the next generation of multimedia search engines, ones that don’t require tagged text.
The Illinois team came home with more than a third-place finish. For professors Hasegawa-Johnson and Huang, Star Challenge was a great opportunity to evaluate their search algorithms.
“As far as I’m concerned I’m very happy and quite proud of the team that we were in the final five, especially that we ranked overall No. 1 in the qualifying round,” Huang said. “That’s the real test of the algorithms.
“The main reason (we did this) is that we have been working on basic research and we always want to see how the algorithms will work in a more realistic setting. In this competition the data is real broadcast TV.”
The opportunity to test his group’s algorithms is also why Hasegawa-Johnson, who has a research line involving audio-visual speech recognition, joined the effort.
“The way they originally framed the task was exactly – at least from an audio point of view – the kind of work that I’ve been trying to move my group toward, that is multi-lingual recognition, language independent speech recognition,” Hasegawa-Johnson said. “And I think similar (to what Tom said) we want to work with realistic, real-world data.
“In fact, the longer term goal is that the information retrieval system we developed for this will be a test-bed for algorithm development for a few years to come.”
Huang said the competition also was good for his students.
“It helps in several different ways,” he said. “One is the students really work together as a team, so I think it was a good experience for them. In trying to solve these problems, along the way not just in the finals, we looked at different algorithms so we learned a lot about what works and what doesn’t work.”
The team also had time for some fun during their trip. Chip Zukoski, former Vice-President for Research at Illinois, is currently serving as Chairman of A*STAR’s Science and Engineering Research Council. Huang said Zukoski had a dinner party for the five teams at Star Challenge and attended the finals.
Despite doing so well in the competition and the experience gained from it, Huang isn’t sure if he would be up for another Star Challenge.
“I don’t know,” he said with a laugh. “It was a lot of investment.”
This article is part of the Winter 2009 Synergy Issue, a publication of the Communications Office of the Beckman Institute.