Reports from the Heart of the Machine
Edit: Zenan
Going to Southeast Asia to promote AI technology will face unprecedented challenges, and they have succeeded.
At the ongoing Beijing Winter Olympics, many black technologies have been moved to the front of the stage: AI referees, visual tracking effects, logistics robots, stir-fry robots, etc., which have become an important part of the competition. With the advancement of technology, artificial intelligence is gradually entering all aspects of life.
When artificial intelligence is mentioned in the technology circle, people will focus more on China and the United States, but ignore Southeast Asia, which has the fastest growth momentum. It has a population of more than 660 million, a fast-growing economy, and a constantly improving infrastructure. With the improvement of consumption levels and the promotion of the epidemic, local technology and talents have not been able to keep up with the rapidly growing business volume, bringing a lot of new demand.
In this market full of opportunities, some AI startups are playing out different stories.
Since its inception in 2016, advance. AI business throughout Southeast Asia such as Indonesia, the Philippines, Malaysia, South Asia such as India, Pakistan, as well as Mexico and Colombia in Latin America, Nigeria in Africa and other countries, and now it is expanding to Europe and the United States.
ADVANCE. AI has successfully served more than 1,000 enterprise customers in the fields of banking, fintech, e-commerce and sharing economy. Despite the severe impact of the epidemic on the global economy since 2020, the company's business has maintained steady and sustained growth, and the number of API calls to cloud service software products has more than doubled, and in 2021, the annual product calls exceeded 1 billion times, which has become a force to be reckoned with in the SOUTHEAST Asian AI circle.
Recently, we partnered with ADVANCE. Dr. Wang Fanglin, head of the AI R&D department, has more than 15 years of experience in the field of artificial intelligence, with a ph.D. from Shanghai Jiao Tong University and a master's and bachelor's degree from Harbin Institute of Technology. He has served as Associate Director of Video Analytics and Machine Learning at NCS Group, Chief Technology Officer of KAI Square, Researcher at the National University of Singapore and Autodesk Research Scientist.
Dr. Fanglin Wang
Fanglin Wang joined ADVANCE in 2018. AI, who has personally experienced the process of rapid business growth, has accumulated rich experience in AI talent team building and technology localization, and shared his views on the common problems encountered by Chinese high-tech enterprises in the development of AI technology in the Southeast Asian market.
AI is a technology but not an industry
From the initial service for China's overseas financial technology companies to the current Southeast Asian local banking, insurance, e-commerce, payment and other multi-field enterprises, with the change of business, ADVANCE. AI's products have also gone through the process from a single function, to customizing solutions according to customer needs, to standardizing and productizing solutions.
As an artificial intelligence service provider dedicated to covering different business scenarios, ADVANCE. AI clearly knows that AI is the way, not the end. "Within R&D, we don't overemphasize AI, and we've always thought of AI as a technology that catches fire at a specific point in time. Every change and restructuring of the team was based on the business strategy at that time, and our aim has always been to help our clients improve efficiency and enhance their business capabilities," said Wang Fanglin.
Compared with some large manufacturers for cutting-edge technology set up AI Lab, ADVANCE. AI's R&D department places more emphasis on the ability to solve real problems, and its challenges are mainly from customer needs. "That doesn't mean we don't go deeper into technology." Compared with the academic community, the practical problems in industry are often more complex and diverse, requiring a sufficient understanding of various algorithms and techniques, and also weighing the boundaries of algorithms and product design."
Currently ADVANCE. The main research directions of AI include computer vision, anti-fraud, credit scoring and other fields, self-developed OCR, in vivo detection and face recognition technology remains leading in the industry, and credit scoring algorithms based on big data and machine learning have won the favor of many Southeast Asian financial institutions.
ADVANCE. Ai proposes a way to model personal information.
Of course, as a research and development department, there are also some people here who will focus on long-term algorithm research, track advanced technologies and propose innovations, hoping to find long-term and more thorough solutions to some core problems.
What is it like to study AI, especially in Southeast Asia such as Singapore, which has long been overlooked? First of all, the talent market here is far less than demand, and the competition for the company is very fierce. In this regard, ADVANCE. AI's strategy is to emphasize the quality of talent, ensure the stability of the team as much as possible, and take advantage of the advantages of multi-country offices to recruit people in Singapore, Beijing, and Jakarta at the same time.
"Startups need to think about target markets in more countries in the early stages. Due to the small size of Singapore's local market, only by facing multiple Southeast Asian countries, and even to China, Europe and the United States can the ceiling of growth be guaranteed to be high enough." Wang Fanglin said. "In the research environment, there are world-renowned university students and research institutions, R&D centers set up by well-known companies at home and abroad, and most of the headquarters of Southeast Asian unicorn companies are also located here, and the overall level of R&D is very high in the general environment." Compared with the domestic culture, it pursues a work-life balance and is very pragmatic."
With the increase in business volume and the continuous accumulation of data and customer service experience, ADVANCE. The technical level of AI has been continuously improved, forming a closed loop of data, technology and products. ADVANCE. AI computer vision products, after more than 10 countries and regions of the market a large amount of real data accumulation, the existing pre-training model and transfer learning methods, can obtain no or very little training data to obtain higher accuracy than competitors.
The diversification of product types has brought advantages to Lingchuang, in the user credit scoring products, benefiting from the internal types of data through, the company can achieve more complete modeling than competitors, largely supplementing the problem that customer data is not rich and does not have modeling capabilities.
From the early days of custom projects, to the SaaS API that can now handle a large number of calls in real time, ADVANCE. From the initial digital identity verification and risk control tools, AI's product categories have developed into a rich product line covering identity authentication, video authentication, and credit scoring. The company's next step will be to propose a series of end-to-end anti-fraud and risk control platform-based solutions.
How to localize AI
SoftBank founder Son Zhengyi has a famous "time machine" theory: the Internet in the United States is more advanced than Japan, first invest in the United States, and then return to Japan with American experience when the time is ripe, as if sitting on a time machine and returning to the United States a few years ago.
In Southeast Asia, which is full of vitality and development prospects, is there a process of "dimensionality reduction" in the research and application of artificial intelligence from China and the United States to India, Indonesia and other countries?
In Wang Fanglin's view, in the short term, China has obvious technological advantages, but the core of the problem is how to truly achieve localization: "No company can maintain the advantages of a certain technology for a long time, and everyone has the opportunity, including local companies." We didn't think of ourselves as a visual or AI company from the start. Work should focus on doing business, and R&D personnel should be clear about how many customers and how much revenue will be brought by investment."
Compared with domestic platforms such as Taobao and JD.com and the APP with face recognition of each bank, the progress of AI technology in Southeast Asia is slower, and the same is true for big data verification and precision marketing. Machine learning applications that have been validated in China have advantages here.
ADVANCE. Ai provides technology that can be quickly integrated into existing applications and dramatically improves the user experience: its algorithm can brush faces, allowing users to match a photo of a face on an account, and no longer need to fill in more accounts and passwords.
ADVANCE. Ai-supplied in vivo detection algorithms.
From 2019 to 2020, Southeast Asia's overall e-commerce retail sales will grow by 50%, far faster than other markets. However, it is not easy to enter this market.
As a developing country market, developing AI technology business in Places like Indonesia will encounter many unprecedented challenges. From a technical point of view, the mobile phones used by people are relatively low-end, low imaging quality and computing power are difficult to ensure user experience, and poor mobile networks will make image transmission delays longer.
ADVANCE. AI must implement localized technical optimization: reduce network bandwidth requirements by compressing models, try to put data processing on the mobile side and use small models; through quality analysis models to prompt image quality problems, and train algorithms to better adapt to local data and solve practical challenges.
In terms of credit scoring, the main challenge faced by data scientists is insufficient data, and the data on strong financial attributes such as transaction flows and credit reports that can be obtained by traditional financial institutions such as banks is relatively difficult to obtain, so the accumulated behavioral data such as lending, e-commerce, operators, and equipment is particularly important, but this has high requirements for feature engineering. ADVANCE. AI technicians use automatic timing feature engineering and deep learning algorithms to extract time series features, which greatly improves the effect and efficiency of feature mining.
In the field of financial credit in Southeast Asia, there are many types of customer groups and huge differences in behavior patterns, and fintech companies cannot use a general credit model to solve the credit scoring needs of different financial institutions, so customer group modeling is an inevitable choice. ADVANCE. AI measures the similarity of customers through unsupervised, supervised learning and other algorithms, so that the customer group changes from a traditional business problem to a technical problem, and finally makes the customer group more accurate and the effect of customer group modeling is more obvious.
On the other hand, technology awareness and regulations also need to be adapted. "A lot of times customers don't understand technology, and in the beginning, we need to spend a lot of time educating customers." Wang Fanglin said.
First important opportunity
ADVANCE. One of AI's biggest customers is a globally renowned bank. When it started its digital identity verification program in 2018, the bank's R&D team didn't have much concept and confidence in how far technologies such as live detection and face recognition could be achieved in Southeast Asia.
"This is our first big international customer in Southeast Asia, with four or five competitors. When we first came into contact, we had more than 20 meetings with Party A, constantly dealing with the senior management with the technical and product teams. Because we are a start-up, it is inevitable that we will be questioned a lot. For example, the first two times I found that their person in charge has always been very impatient." Wang Fanglin introduced.
Fortunately, the bank's supplier research has been carried out dutifully.
In the process of advancing, ADVANCE. AI is constantly improving its own technology, they simulate the actual experience of the end user, complete the beta APP on the mobile phone, let the person in charge directly experience, demonstrate the convenience of the new technology and the solution of abnormal situations.
ADVANCE. AI is finally getting the chance to move on to the next round of validation. "Our products are more in line with Southeast Asian characteristics, and in the end, it is easier to beat all competitors in the accuracy testing stage."
But this is just the beginning, after winning the bid, there is a long process of technology landing and compliance, which eventually took a total of more than ten months. Before the solution went live, the project went through three internal testing, demonstrations, and pre-launch processes. ADVANCE. Ai's teams also need to help banks select hardware for deep learning deployments and provide advice on infrastructure.
Multinational companies place great emphasis on data privacy and compliance, with a focus on internal processes and norms, including procurement processes and contract terms. "Customers will ask us to guarantee that there will be no reputation problems in the next few years, and we will have to make compensation in the event of a breach of contract," Wang said. "The AI models we implement can't discriminate across races and genders. In terms of safety and reliability, they conduct research on various technical aspects of the product to determine which technologies are used."
In the process of continuous communication, we can find that these overseas companies can be relatively objective, and from a technical point of view, the competition in emerging markets is not as fierce as in China. This presents an opportunity for AI startups.
Products are the biggest demand
Expanding the team after determining the business outlook and choosing the direction of earnings in the coming years may be the right way for tech companies to build AI Labs.
As the business continues to expand, ADVANCE. AI's team size has expanded from a few dozen people to nearly 200 people today, and the office has grown from two countries to five countries.
Now, ADVANCE. AI's R&D team consists of engineering, computer vision, data science, and a data annotation team, and the company is also developing its own machine learning platform.
Since its inception, ADVANCE. On September 23, Link Group announced that it had completed more than $400 million in Series D financing, led by SoftBank Vision Fund Phase II and Warburg Pincus Investment, followed by Polaris Capital, Yuanjing Capital, Gao Rong Capital and Singapore Economic Development Board Investment. After the completion of the financing, Lingchuang Group has been valued at more than US$2 billion, becoming one of the largest independent technology startups in Singapore.
Facing the future, ADVANCE. AI will still stick to its own path. "Personally, I prefer to think big, start small, and try to be down-to-earth. We need to make quick trial and error in business, quickly validate technically, and concentrate our firepower in the direction of firm confirmation." Wang Fanglin said.