Techworld Interview – Overcoming Challenges at Alchera Inc.
“Securing data is a challenge for us, too, but overcoming it makes us competitive. On the other hand, companies that are unfamiliar with the data do not have this competitive edge.” – Young Hwang, CTO and Co-founder, Alchera Inc.
Made of those who view challenges as the foundation for significant growth, Alchera is an AI startup established in 2016 by two talented people from Samsung Advanced Institute of Technology.
Alchera’s growth is remarkable. At the same time as its foundation, it signed an investment contract with Naver to supply face recognition engines to the famous camera app SNOW. As an early startup, it won a project to develop an immigration system at Incheon International Airport and achieved a feat by beating China’s AI unicorns at an international competition. On top of that, Alchera’s growth rate, which is set to be listed on the KOSDAQ this year seems unusually fast compared to other startups.
What is the driving force behind their success? We met with founders Jay Kim and Young Hwang at their headquarters located in Pangyo Techno Valley to talk about Alchera’s technology and vision.
Alchera is a company that has strong AI-based face recognition. In particular, the FRVT 2018-2019 attracted attention by beating out Chinese AI unicorns. What is Alchera’s core technology competitiveness?
The key competitiveness of Alchera’s face recognition engine lies in its sophisticated eye, nose, and mouth identification capabilities. The perceived error range is only about 1 mm. In international face recognition technology competitions, such as FRVT (Face Recognition Vendor Test), not only did we get good reviews from the front, but also in poor perception environments where subjects are occluded, or there is poor lighting or unusual camera angles. In particular, database test results showed a performance that can be determined up to the 17th year.
Does this mean you can also check if the subjects in the past pictures match?
Yes, this ageing reading technique can determine the actual match between the owner and the ID photo, which is never updated for an extended period of time once taken. The automated immigration system at Incheon International Airport, developed by Alchera, is also applied with this aging identification function.
We would like to emphasize that this is proprietary technology. Currently, only six to seven companies have engine development technology in Korea. Except for them, most of them are developing products with overseas engines, which makes it difficult to compete with global companies. When foreign companies, including Japan’s NEC, a strong player in the face recognition field, expanded their service in Korea, we have been able to refine and refine the original technology without wavering. And I think this has been a big help in securing Alchera’s unrivaled competitiveness.
What other businesses are there using face recognition technology?
In partnership with Shinhan Bank in the fintech sector, the company launched Shinhan FacePay, which allows users to pay with face information. Also, it is working with various financial companies.
For example, the process of ID authentication to open a bank account initially required a resident registration card or a driver’s license. But we are also making passports available here. Let’s think about it. Even minors have the right to open a bank account legally. But these guys don’t have any ID yet, do they? On the other hand, passports can be made by anyone regardless of age. In other words, more people will be able to enjoy the benefits of the non-face service once it becomes available through passports. We continue to think about improving accessibility using face recognition technology.
There are concerns that face recognition is relatively less secure. Was there a problem with that?
In fact, when it comes to security, iris recognition is the safest, followed by fingerprint and then face recognition. However the security gap between the fingerprint and face is decreasing. In addition, since the face is non-contact, it is much easier to use, more hygienic, and has more potential use cases. Besides, the face is not information that disappears. Some people may have their fingerprints worn out, but their faces are the ones who stay with the user until the moment they die.
In the case of the iris, security is high, but the recognition process is difficult.
In addition, Asians with relatively small eyes have a lower iris recognition rate than Westerners. Of course, it is much more efficient to use face recognition security if one can guarantee sufficient security with just his face.
Alchera has been building a stable growth base by attracting investments from various conglomerates and VCs since its inception. Money power, technology, all feel more relaxed than regular start-ups, but is there any specific business that you have attempted based on that?
Recently, We invested in developing an abnormal situation detection technology called Visual Anomaly Detection Technology (VADT) to develop solutions specialized for fire monitoring. Usually, forest fires have golden times 20 minutes after ignition. The problem is that 20 minutes of smoke is almost a small amount, so it is not easy to distinguish between fog, clouds, haze, etc. with the naked eye. Alchera’s AI solution made it possible to respond early. Perhaps the situation would have changed significantly if such a detection system had been in place in time before the last Australian forest fire accident. In domestic and international cases, Korea Electric Power Corp(KEPCO)has already introduced 240 cameras to monitor fires and California has also introduced 200 remote forest fire surveillance cameras. In addition, the company is equipped with technology to diagnose changes in various facilities in real-time.
For AI companies, data is a key competitive edge. Collecting is important, but the process of refining it to fit the use is even more important. What efforts are you making to strengthen your data competitiveness?
In Alchera’s case, how fast we can create learning data has been outstanding since the beginning of your business. We have developed our own data labeling tool, and about 100 workers are currently being employed in data collection and refining operations here in Korea and Vietnam. It’s also cloud-based, so hundreds of people at the same time have no problems processing data.
In fact, compared with China, where the country collects data directly and provides it to companies, Korean AI startups have thought that they would be relatively disadvantaged in the competition to secure data. Still, what’s the secret that Alchera was able to beat China’s AI unicorns in the technological race?
We saw the environment as an opportunity, which we say is disadvantageous. Although Chinese companies are clearly at an advantageous position in terms of securing data, what is more important than quantity is quality. We have invested early in research to achieve high accuracy with as little data as possible. In addition, we have the know-how to efficiently categorize what data we need in the data we already have.
In situations where data is difficult to obtain, sometimes you create it yourself. Recently, we have produced about 1.25 million virtual forest fire data and trained them in the system, resulting in good results. Securing data is definitely a challenge for us, but I think overcoming it makes us competitive. On the other hand, Chinese companies that are familiar with the data given do not have this competitive edge.
How is your overseas business? In particular, technology trends in Southeast Asia such as Vietnam, where Alchera Corporation has entered, are somewhat unfamiliar.
Currently, Ho Chi Minh City in Vietnam is mainly engaged in data processing. Jay visits the country once every two months to meet with government agencies or Vietnamese businessmen to introduce the company and search for cooperation points. The level of technology in Southeast Asian countries is not that high. I think the benchmarking for successful technologies is stronger than self-development. Services such as Uber and food delivery are already prevalent in Southeast Asia, and demand for them is exploding. Therefore, we are also considering the direction of cooperation with local Vietnamese companies.
I heard that they are also interested in software as a service (SaaS) business beyond engine supply.
We have newly signed a SaaS-based partnership with Kajima, Japan’s leading construction company. The company is in Japan, but the engine of the system they introduced is in Korea, and updates are made in Korea.
The deal is symbolic in that it is SaaS, the first face recognition field in which a Korean startup has succeeded in entering foreign markets. It is also meaningful in that Alchera has made a new leap from a technology development company to a company that also supplies services.
While the current diplomatic relationship with Japan is troubling, we are not just complacent about the deal, but we are determined to lay the groundwork for more of our products to expand abroad.
Face recognition-based services are always followed by privacy infringement issues. How is Alchera responding to this problem?
We are legally paying for the tens of millions of image data needed for learning. All of them have received notified consent forms, and related agencies such as the Ministry of Justice have confirmed that there is no problem with them.
The company does not save face information that is obtained by operating Face Pay. For the sake of understanding, it is not about saving the image of the face itself, but by extracting only the mathematical features from the front. Since this information cannot be reinterpreted, it can ensure safety in terms of privacy protection. And because even this feature data is stored in a physically dispersed environment, we cannot easily access it either.
I especially feel that I emphasize legitimacy.
Not merely to avoid sanctions, but because legitimacy is also one of our competitiveness. The data sources of Chinese companies competing in this field are in fact, hard to see as clean. And the problem becomes a drag on these companies when they leave China and go abroad. No company would like data from a trading company if it had a criminal element. In this sense, Alchera, who conducts business under a thorough legal notation, can become a relatively reliable business partner.
Fortunately, recent policy issues for collecting personal information have been being resolved little by little. Unilaterally opening up data is not a democratic way. The policymakers above are doing well now, but it is hard to ignore anyone since their stakeholders are the people. These are the parts that need to take time to reach sufficient social consensus.
There are some notable early achievements, such as the contract with Snow, but such a business will not last forever. In addition, a variety of services using AI image recognition technology are appearing It seems like it’s time for Alchera to prepare for his next target. What are the current prospects for this field?
First, there may be aspects of expansion in fintech and the public sector, and it may be used in-store management. Closely, LetinAR, which collaborated with Alchera at MWC 2019 and CES 2020, is a good example.
LetinAR is a startup with global top-level AR optical lens production
technology. In the last exhibition, we applied our technology to AR glass, and if you look at a person standing at a booth, you can reinforce that person’s information and get positive reviews.
When this comes to reality, employees will be able to provide better services at offline stores at any time based on information and preference data from customers’ previous visits. Also, Alchera’s face recognition engine is highly competitive when compared to IBM or AWS. This is due to our engine being light-built for Snow app-based mobile in the first place.
What are your future plans and expectations?
First of all, there is a plan to list the KOSDAQ soon, and we expect it to be completed within this year. It will also strengthen Alchera’s business base and funds. The hope is the addition of more talent.
I also think that it is time for us to build global recognition, as we are quite well known in Korea. We hope that more talented and customers will come from abroad to promote our technology more widely.