Alibaba, XPeng Build China’s Largest Computing Center for Autonomous Driving; Pony.ai Sues Autonomous Truck Startup; Huawei Teams up With TCM Giant to Design Drugs Using AI

Weekly China AI News from Aug 1 to Aug 7

News of the Week

Alibaba, XPeng Build Mega Computing Center to Train Self-Driving Models

What’s new: China’s largest cloud service provider Alibaba Cloud and electric vehicle upstart XPeng launched a mega computing center to accelerate the training of AI models for autonomous driving software. Named Fuyao, the Inner Mongolia-based computing center put together GPU clusters that provide a computing capacity of 600PFLOPS, boosting the speed of XPeng’s current training by 170 times. Xpeng said the Fuyao computing center has been employed to train the model used in XPeng’s advanced driver-assistance system (ADAS) named NGP (Navigation Guided Pilot), which is available on highways and will expand to urban environments by 2022.

Why it matters: As self-driving tech becomes a striking advantage to stand out amongst electric vehicle (EV) competitors, EV companies are tapping powerful computers to process a massive trove of collected data for autonomous driving software. Said XPeng CEO He Xiaopeng, in the next three years, XPeng will face a computing power demand that is 10–100 times higher than the one today. Tesla also revealed its supercomputer named DOJO that will reach 1.8 exaFLOPS or 1,800 petaFLOPS. Unlike Tesla’s in-house development, XPeng relies on Alibaba, a key investor of XPeng and former employer of CEO He, for computing supply.

Why is Inner Mongolia? Ulanqab, a city in Inner Mongolia of north China, is the home to Fuyao along with other data centers of Huawei, Alibaba, Apple, and many other tech companies in China. The city has a favorable climate and abundant energy sources — its temperatures can drop to -22 degrees Celsius in winter as a perfect free-air cooling system. Inner Mongolia is one of the eight regions where China is deploying mega data centers under its infrastructure plan “East Data West Computing”.

Pony.ai Sues Autonomous Truck Startup Founded by Ex-Employees

What’s new: Chinese autonomous driving upstart Pony.ai has sued a 9-month-old autonomous truck startup and the firm’s two founders, who are also former employees of Pony.ai, over alleged trade secret infringement. In a lawsuit filed in the Beijing court, Pony.ai is seeking compensation of RMB60 million from Qingtian Truck and requesting the company not to infringe its business secrets anymore.

In an official statement responding to the local media, Qingtian Truck said the company has not infringed any third-party trade secrets and has been regulatory compliant.

What’s Qingtian Truck: Qingtian Truck was founded by CEO Frank (Zhenhao) Pan, the former CTO of Pony.ai’s truck business PonyTron, and CTO Youhan Sun, the former US head of planning and control at PonyTron. The robotruck company raised $10 million in its Seed Round this January led by 5Y Capital, which also happens to be a key investor of Pony.ai.

Waymo vs Uber, Baidu vs Jingchi: The Pony.ai lawsuit echoes the 2017 Waymo vs Uber dispute in which Waymo accused Uber of stealing its LiDar technology and the 2018 Baidu vs Jingchi lawsuit (now known as WeRide.ai) over a similar trade secret infringement complaint. Waymo and Uber later settled their dispute after Uber paid $245 million worth of company shares, and Baidu and Jingchi also buried the hatchet after CEO Wang Jing departed and Jingchi joined Baidu’s partnership programs.

Huawei Teams up With TCM Giant to Design Drugs Using AI

What’s new: Chinese major pharmaceutical company Yunan Baiyao Group — a traditional Chinese medicine (TCM) household name — announced that the company has signed a three-year cooperation agreement with Huawei on “AI Drug Research and Development”. The two parties will work together and tap AI on small and large molecule design, relevant disease treatment, and database development.

Fueled by Huawei’s AI capability, Yunnan Baiyao plans to set up an R&D team to apply AI in molecular designs, drug target research, immune systems, TCM syndrome research, and image recognition.

Who’s Yunan Baiyao? Yunnan Baiyao is a household Chinese herb medicine to curtail bleeding and has a history of over one century. It supports circulation in the body and relieves minor aches and pains. The drug is manufactured and distributed by the Yunnan Baiyao group, which is government-owned in Yunnan, China.

Huawei’s drug molecule model: In 2021, Huawei co-developed and -trained its Pangu drug model with the Shanghai Institute of Materia Medica, Chinese Academy of Sciences. The model can appropriately characterize the molecule from both representations and improve the performance of downstream drug discovery tasks. After pre-trained with 1.7 billion small molecules, the model achieved state-of-the-art results in 20 drug discovery tasks in the market. A Pangu molecule generator also built up a library of 100 million novel drug-like small molecules, with a structural novelty of 99.68%. You can find the paper PanGu Drug Model: Learn a Molecule Like a Human on Biorxiv.

Why it matters: Huawei’s partnership with Yunan Baiyao reflects a rising ascent of AI-power drug discovery in China. We all know drug discovery is costly and time-consuming. Pharm companies could cost over 1 billion dollars and take over a decade to put a new drug on the market. DeepMind AlphaFold’s success in protein structure prediction depicts a bright future where deep learning will be able to shorten the drug discovery process and design novel drugs for untreatable diseases. Chinese AI and tech companies like Alibaba, Tencent, Huawei, and Baidu have quickly dived into this new marketplace.

Papers & Projects

Ekko: A Large-Scale Deep Learning Recommender System with Low-Latency Model Update

Deep Learning Recommender Systems (DLRS), which are widely adopted in many web applications like Facebook, Amazon, and TikTok, need to update models at low latency to suffice new users. However, most existing DLRSs like Merlin, Google’s TFRA, and Meta BigGraph train and validated models offline and then broadcast updated models to new users, leading to a significant latency. Thus, researchers from Tencent and the University of Edinburgh proposed Ekko, a novel DLRS that enables low-latency model updates. Ekko can reduce the model update latency to 2.4 seconds. The paper has been accepted at OSDI: Operating Systems Design and Implementation.

HelixFold-Single: MSA-free Protein Structure Prediction by Using Protein Language Model as an Alternative

Researchers from Baidu and BioMap proposed HelixFold-Single to explore the limits of fast protein structure prediction by using only primary protein structures. HelixFold-Single combines combine a large-scale protein language model with the superior geometric learning capability of AlphaFold2. The code of HelixFold-Single is available at this URL, and a stable web service is provided on this URL.

Rising Startups

IM Motors, a smart electric vehicle (EV) brand backed by China’s largest car company SAIC and e-commerce giant Alibaba, has raised RMB3 billion in its first public financing. Founded in 2020, the company has delivered over 1,000 L7, the company’s first luxury sedan that starts at RMB 408,800, since June.

Fosun Aitrox, Fosun’s first incubated artificial intelligence company for medical diagnosis, has raised over RMB100 million in a strategic funding round. Founded in 2018, the Shanghai-based company focuses on the research and development, production, and sales of medical imaging and pathological software for early screening and diagnosis.