HK experts note progress in face recognition AI
06 February 2018 16:45 GMT

AI-based facial recognition technology has advanced drastically to achieve a negligible deviation after outperforming human eyes for the first time in 2014.

Xiaogang Wang, managing director of SenseTime Research Institute revealed the advances this week, DigiTimes reported.

SenseTime also works with the Ministry of Public Security's Third Research Institute, in Hong Kong, to develop the mainland's citizen network identity infrastructure.

The progress is equivalent to the evolution from a 4-digit password to an 8-digit one within four years, Wang said, while delivering a keynote speech on the topic of "Now and Future under AI Empowerment" at the EmTech China conference hosted by MIT Technology Review in late January in Beijing in his capacity as head of the research arm of China's AI unicorn SenseTime.

Wang said the most outstanding achievement in the rapid AI (artificial intelligence) development in the past decade rests with deep learning, stressing that there are three major factors driving the development of deep learning: big data, cloud computing powered by GPU and AI chips, and constant algorithm innovations.

He said that constant upgrades of algorithms have sharply boosted the AI-based face identification capability through powerful neural networks that have developed from 5-layer architectures to 1,200-layer ones to become capable of accurately identifying suspects targeted by law enforcers from among hundreds of thousands or even more facial images.

The ever-advancing algorithms can make cities safer, Wang said, adding that some criminal suspects wanted by the police for years were finally arrested as their faces can be easily captured and recognized through the dynamic face distributed control system despite their changing names and using fake ID cards. The facial recognition technology can also help locate missing elderly people or children.

So far, Wang disclosed, SensTime's "face unlock" technology has been applied to over 100 million smartphones through the support of Qualcomm chipsets.