Better understanding of facial recognition needed for reasonable social dialogue, says YITU Technologies AI research scientist
YITU Technologies has revealed to Biometric Update that it is expanding its Silicon Valley research lab, which will be managed by YITU AI Research Scientist Dr. Wu Shuang, as the company continues to raise its international profile.
Dr. Wu told Biometric Update in an exclusive interview that the company has a wide range of applications of its convolutional neural network technology in various stages of development. It is getting positive early feedback from its facial recognition system for cardless ATM transactions, for example, but its deployment still presents technical challenges.
The system is currently deployed to thousands of ATMs, and the scale, along with the inherent sensitivity of the application, add significant complexity to its requirements.
“There are technical challenges. Because it is widely deployed, the environment is less controlled,” Dr. Wu explains. “We have lighting issues, or sometimes the cameras are slightly fish-eyed, and also you need to have liveness detection.”
In healthcare, YITU is taking similar AI imaging technology to that used in its facial recognition solutions, and using it to for medical diagnosis tools, such as performing early screening for signs of lung cancer.
“The fundamental tools are the same, but you are using it in different scenarios,” Dr. Wu says. “The data is different, but the requirements are also different in terms of recall and precision.”
While it already counts customers beyond China throughout South East Asia, the Middle East, and Africa, YITU is reaching out to international markets, as it has planned to all along, according to Dr. Wu, with an office opened early this year in Singapore, and staff currently active in Europe.
The international profile of YITU was raised by the company’s win of the NIST Face Recognition Vendor Test (FRVT) in September Dr. Wu says, which it followed by winning the IARPA Face Recognition Prize Challenge (FRPC).
When asked about controversy arising recently from the relatively lackluster performance of leading facial recognition systems at identifying women and people with darker skin, Dr. Wu notes that not only is the problem a natural product of the situation, it has also been well-known within the industry for a relatively long time.
“That means as a professional face recognition service provider, you must be aware of that, and also do things to make sure you test it thoroughly, in such a way that you know, even in challenging cases, you will have good results,” he says. “Between you and your customer, you have to have full disclosure. That means not just a single number for each metric.”
He notes that in many cases in which the industry is communicating about accuracy, such as the NIST FRVT, the score is given with breakdowns according to race and gender, in recognition of the need to acknowledge and reduce bias in systems. Further, according to Dr. Wu, YITU’s technology shows much lower variation in accuracy between populations than those commonly cited in the media.
“We have seen some articles floating around with catchy titles, that we don’t think explain where we are correctly,” Dr. Wu observes. “These are not helping a reasonable public debate on this issue, and that’s not going to help the whole industry moving forward.”
With the rapid advance of facial recognition technology, and the increase in its popularity, Dr. Wu says that real technical and ethical issues need to be addressed. For that to happen effectively, however, the level of education about AI and facial recognition technology must significantly improve.
“I believe facial recognition has improved a lot in recent years,” Dr. Wu says. “It is still not widely understood how much better we have gotten, even in the industry. The industry and general public have to be informed about this rapid progress, and then we can talk about how to apply such advanced technology in a more reasonable way. We need to establish industry standards, and also policy-makers need to set up policies and regulations for facial recognition to be used properly and widely, and to void the ambiguities and doubt about how it can be used.”