NTU team breaks captcha system for global award

Students' deep-learning network gets perfect accuracy score

(From left) co-adviser Lu Shengliang, team members Liu Siyuan, Tan Ying Hao, Ren Daxuan, Li Yuanrui, Shao Yiyang and Wu Bian, with Prof Francis Lee, also an adviser, at the International
Supercomputing Conference 2017 in Frankfurt, Germany.
PHOTO: NATIONAL SUPERCOMPUTING CENTRE SINGAPORE/NATIONAL TECHNOLOGICAL UNIVERSITY

Breaking a captcha system may result in serious privacy and security implications, but for six undergraduates, doing so spelt a breakthrough.

A captcha is a computer program or system which can tell the difference between human and machine input, and is generally used to prevent spam and automated extraction of data from websites.

At the International Supercomputing Conference held in Frankfurt, Germany, last month, a team of six from Nanyang Technological University (NTU) attained a perfect accuracy score in the captcha component of the student cluster competition. The team, sponsored by the National Supercomputing Centre Singapore (NSCC), achieved the centre's best showing yet at an international high-performance computing conference.

A captcha - usually a distorted string of letters and numbers - is a common authentication method that adds an additional layer of security for websites.

The 11 student teams from top universities, including the Massachusetts Institute of Technology and Harvard, had to develop a network that could recognise and read the letters in captcha images. Each team was given training images to build a model with its computing cluster. Following that, unseen images were tested on the model.

The NTU team was given the Deep Learning Award for their stellar performance in that component. Deep learning is a new area of machine-learning research that involves feeding a computer system a lot of data, which it can use to make decisions about other data.

This method of machine learning is behind some popular artificial intelligence applications like image recognition and self-driving cars.

The honour accorded to the NTU team was shared with overall first and third place winners Tsinghua and Beihang universities respectively, which also attained perfect accuracy scores. The other two universities are from China, where the world's top two fastest supercomputers reside. In comparison, Singapore's first petascale supercomputer Aspire 1 ranks 136th.

"This achievement has shown that Singapore is on track towards building strong high-performance computing competencies among our institutions of higher learning," said Associate Professor Tan Tin Wee, the executive director of NSCC.

NSCC, together with the Agency for Science, Technology and Research (A*Star) Computational Resource Centre, has been training students for such competitions since 2014, and the NTU team took part in last year's edition of the competition.

Team captain Liu Siyuan, 21, a third-year student from the School of Computer Science and Engineering, said: "We are very excited to have won especially because all teams were quite strong."

The team will be heading to the United States in November to compete in SC17, the International Conference for High Performance Computing, Networking, Storage and Analysis.

Join ST's WhatsApp Channel and get the latest news and must-reads.

A version of this article appeared in the print edition of The Straits Times on July 14, 2017, with the headline NTU team breaks captcha system for global award. Subscribe