China’s ridesharing unicorn DiDi Chuxing unveiled its “DiDi AI Labs” today in Beijing. The new AI lab expands on the DiDi Research Institute and DiDi Labs, and will cluster over 200 AI scientists and engineers to “improve user experience and transportation efficiency and further enhance [the] intelligent transportation ecosystem,” according to a company press release. Didi AI Labs will be led by company Vice President Professor Jieping Ye.
DiDi manages some 20 million daily rides and serves 400 million active users. The company sits on an impressive valuation of US $56 billion, and in 2016 bought Uber China’s operation for US $7 billion, further consolidating its domestic rideshare monopoly.
DiDi Labs focuses on big data and smart traffic and conducts AI research on NLP, computer vision for operations research, deep learning, statistics, etc. Research cases include predicting user destination, suggesting pick-up points, intelligent order dispatching, estimated time of arrival (ETA), and routing optimization.
AI Labs head Jieping Ye is a University of Michigan professor and a specialist in machine learning, data mining and analytics, and specializing large-scale sparse model learning. In an April 2017 lecture at Peking University, Prof. Ye said DiDi had accumulated so much data that “if you do the model, sample size can easily go up to billions.” DiDi Rideshare vehicles submit updated GPS data every few seconds, processing 2000 terabytes each day.
At the lab launch DiDi also showcased its “DiDi Smart Transportation Brain”, a tech solution for smart city traffic management integrated with local traffic authorities that has already been adopted by more than 20 Chinese cities. Smart Traffic Signals installed at 344 road intersections in the city of Jinan have saved 30,000 hours of travel time and shortened traffic delays by 10–20%. The system is now deployed at 1,200 intersections nationwide.
DiDi Labs is becoming increasingly active and visible in global AI. Last year it published the paper A Taxi Order Dispatch Model based On Combinatorial Optimization at leading international data conference KDD, participated in computer vision conference CVPR, and was a NIPS Platinum Sponsor.
Journalist: Meghan Han | Editor: Michael Sarazen