On my Streicker Fellowship this summer, I had the opportunity to do computer vision research at an early stage autonomous driving startup in Beijing called Momenta. I was able to study the edges of a topic directly related with my upcoming thesis while working in an environment which enabled me to gain a much better understanding of the state of the Chinese tech scene, some of its unique characteristics, and the many ways in which Beijing is poised to steal the show in the upcoming AI revolution.
I researched realtime video segmentation with Momenta's freespace segmentation team. The ability to determine the visual region in which it is safe to drive will be a vital skill for self-driving cars, and as a consequnce, our work influenced the development of many higher-level path-planning and driving models. Below is an example of freespace segmentation performed by the Israeli company Mobileye.
This can be a tough task! Especially since we wish this to be failsafe in complex and novelscenes, and we also must determine the types of objects adjacent to the drivable space, but this can all be done using variations on a model called a "convolutional neural network." The models are trained using millions of miles worth of data which we have crowdsourced from taxis and private cars all over China, and to train the models, we need to use the fast computing power of GPUs. Because of this, Momenta must maintain a large number of these units (shown below) for the use of its research divisions.
Before being curried away to our models to train, we use a number of data auditing and filtering techniques. Most of the pictures have to be annotated by hand. With so many cars driving around the chaotic streets of China, it is not uncommon to come across comedic scenes like that captured a week ago (pictured below).
In order to compare our technology to the state of the art, we also sought to test drive competitors' vehicles, looking through the debugging information while getting a qualitative gauge of the areas in which we have the advantage. One such competitor is Musk's Tesla.
Momenta provided me with a wonderful community of coworkers of all ages. One of my new friends there learned that I teach swimming lessons at Princeton, and so we arranged for a pool day with her family, another new friend Xu, and her parents, who were visiting for the weekend. It was wonderful to take part in a family outing and even more rewarding to speak for hours with Xu's parents at dinner to hear their take on life as residents of one of China's second-tier cities.
Another summer goal was to gain a better understanding of the types of entrepreneurial opportunities in Beijing and China. I left this summer with a more nuanced understanding of the ways in which AI and other high-tech tools may be employed to power social enterprises. I served as a grant committe officer for Beijing-based startup Philanthropy in Motion through their model foundation program. Through this program, I connected with the environmental social enterprise Smart Air to learn more about their work, conduct due diligence, and pitch their operation to the committe to compete for a venture fund of 20,000 USD. Smart Air is an organization with the mission of letting everyone breath clean air. In addition to selling filters, it prioritizes education as a tool for individual empowerment, teaching citizens of over 4 countries of the importance of clean air and the relatively cheap ways in which individuals may create their own filters in order to safeguard their lives and the lives of their loved ones. To anyone wishing to live in Beijing or any other developing area plagued by high levels of air pollution, I would highly recommend you check out their website to learn more about how you can make an affordable air filter.
At model foundation, we met with grassroots organizers, heads of large foundations such as the Ford Foundation and Ai You, and dialogued with entrepreneurs and leaders from all across China who wish to make a lasting impact on a wide variety of pressing issues.
Apart from the conference, I also sought to visit other high-tech startups and companies to learn more about the Zhongguancun tech ecosystem. Aside from a number of conferences and visits, I had the opportunity to visit Google China (below) and meet the founder and former CEO of AutoNavi, a leading Chinese mapping company (for instance, their data powers Apple Maps in China).
Beijing wasn't all work, however! One of the reasons I was excited and fortunate to work at Momenta is that the founding team all had previously been members of Tsinghua's outdoors club. My group decided to camp out one night to explore the mountains and woodlands 200 km north of Beijing city. While our initial plans went sour due to inclement weather and the realization that access to certain areas was restricted to foreigners, we still were able to see much of the natural wildlife of norther China, explore the grasslands, and witness the large-scale wind farms the CCP has installed across this region.
And of course, my culinary adventures in Beijing were spurred on by a group of fellow interns who made it their goal to have me taste of all of China's specialties by the end of the summer. While I only tried a small subset of the gastronomical managerie on which country dines, I did end up trying some various delicacies such as duck tongue and pig brain (shown below).
These are only a few snapshots from a summer of constant immersion and growth. My Streicker fellowship gave me learning opportunities that couldn't have been obtained elsewhere. Thank you Mr. Streicker, and thank you Momenta for giving me the once in a lifetime opportunity to research alongside such incredible people!
Will is one of 12 undergraduates who were awarded 2017 Streicker International Fellowships. Streicker Fellows design their own projects or internships in conjunction with a hosting organization. During his fellowship, he worked at Momenta, an artificial intelligence startup in Beijing, where he will research computer vision algorithms for autonomous vehicles and study the policies of the new technology's implementation in different social contexts.