If there would be a league for buzz words, Artificial Intelligence would certainly be a permanent resident in the top ranks. In 2017 alone, the industry received over $15.2 billion dollars in global funding, up from just $1.7bn in 2013. Consistent with the funding explosion in the Chinese market, China attracted 48% of capital invested globally in AI in 2017, dethroning the U.S. with a share of 38% (yes, you counted right, that leaves a meager 14% for the rest of the world). To highlight the massive change, China’s share in global AI funding in 2016 was just slightly above 11%. Considering Chinese funding dominance in AI, is it no wonder the most valuable AI startup globally is a Chinese one. Raising its Series C and Series D in May and September 2018, SenseTime is currently valued at $6bn with a total funding amount of $2.6bn.
What is SenseTime?
Founded in October 2014 by a team around Xiao’ou Tang, information engineering professor at the Chinese University of Hong Kong (CUHK), SenseTime is a company at the heart of AI research and application in multiple fields, ranging from facial recognition to autonomous driving.
Just slightly over 4 years old, the company underwent a stellar ascent to rival tech heavy-weights such as Google and Facebook in AI research. The company’s revenue grew over 400% annually over the last 3 years and became profitable in 2017. In this short period of time, SenseTime grew to over 2200 employees. Its workforce comprises over 150 AI PhDs, propelling the company’s joint research lab to one of the top 10 AI labs globally. The number of PhDs SenseTime was able to attract in this short amount of time is specifically impressive given the fierce competition for AI talent globally.
The startup is connected to over 700 customers, partners and alliances. Its most notorious customer is the Chinese government, which uses SenseTime’s technology as a backbone of its national camera surveillance system. Its partnerships are formed to make technological advancements in fields such as mobile and IoT solutions (Qualcomm) or autonomous driving (Honda). SenseTime’s major competitor in the Chinese market is Megvii (aka Face++), which is currently raising $500m at a $3.5bn valuation. Face++ has become a global heavy-weight in facial recognition as well but relies on open-source technology of third parties.
It is worth mentioning that against the common perception of Western media, SenseTime is not a Hong Kong-based company, but incorporated under Chinese laws, headquartered in Beijing (even Wikipedia is wrong sometimes).
SenseTime was founded to commercialize the facial recognition algorithm DeepID, achieving a 99.45% recognition rate on the Labeled Faces in the Wild (LFW) database. The algorithm was first to surpass the accuracy of human eyes, outpacing Facebook’s previously leading technology. Facial recognition is core to SenseTime’s business, serving the government in analyzing footage from its 170mn CCTV cameras. The company leverages its facial recognition capabilities in other use cases as well. The technology was used by its customers for 400mn facial identity authentications in 2016 alone. Earlier this year, SenseTime rolled out a pilot project in Zhangzhou, a 9.4mn city, equipping police with smart glasses that identify potential suspects by matching the footage generated on the ground to a government database. The company claims to work individually with more than 40 Chinese cities. Part of the funds SenseTime raised in 2018 will be dedicated to further develop its “Viper” algorithm to analyze 100,000 data feeds (camera footages) simultaneously up from its current capacity of a couple thousand. The algorithm is used for camera footage analysis and is core to SenseTime’s push to champion mass-surveillance technology.
The company’s further product offerings include, among others: SenseDrive - a range of autonomous driving solutions, SensePhoto – a mobile image processing solution, SenseAR – an augmented reality rendering engine designed to identify objects, persons and track gestures and SenseGo – a business intelligence solution to analyze consumer behavior from video footage. The company is furthermore providing security software and forging partnerships to develop smart city solutions.
The CUHK-SenseTime Joint Lab published over 400 computer vision-related papers. The following graph shows the number of publications of global AI research powerhouses between 2015 and 2017. The publications displayed are limited to the research receiving acknowledgement by the most distinguished AI conferences globally.
While its competitors rely on open-source technology, SenseTime built a proprietary AI platform. SenseParrots matches industry-leading supercomputing platforms such as TensorFlow (Google) and Torch (Facebook). It allows SenseTime to better tailor the technology to specific research purposes and use cases, such as optimizing the platform to train super-deep neural networks in image recognition. This way the company created a neural network of the unprecedented depth of 1,207 layers in 2016. This neural network can train on 2 billion images simultaneously. To put this in perspective, Microsoft’s ResNet, the deepest, trained neural network for image recognition in 2015 had a depth of 152 layers. While the quality of a neural network depends on a multitude of variables, depth coupled with large amounts of data increases the algorithm’s accuracy. Earlier this year, SenseTime reported that it built 3 super computer clusters comprising 8,000 GPUs in Beijing as well as one cluster in Hong Kong. This hardware upgrade aims to further increase the speed of training algorithms on billions of data points and thus developing new applications even faster.
SenseTime's Future Aspirations
Going forward, SenseTime envisions fueling a multitude of industries with AI applications. It is currently eyeing 18 industries ranging from established fields such as text recognition to emerging fields such as AI-powered education and augmented-reality advertising. SenseTime intends to use the $1.6bn raised in 2018 to finance further investments into adjacent fields such as autonomous driving, healthcare, IoT and augmented reality. The money will furthermore be used to increase computational capacities as well as keep up with the exploding cost for AI talent.
The competitive advantage SenseTime is building up over time by accessing billions of data points collected by the Chinese government is substantial. This data treasure chest will help the company to develop best-in-class algorithms and software solutions to continue to rival industry leaders. The AI industry is estimated to grow 42.5x from slightly over $4bn in 2016 to almost $170bn by 2025. SenseTime, thus, sensed the right time to make its rapid ascent to an industry leader and future will tell if the company can perpetuate its stellar performance in these exciting times.