Suggestions
Tianshi Gao
Distinguished Engineer at Cruise
Tianshi Gao is a Distinguished Engineer at Cruise, where he has been working since May 2019.2 He is a machine learning expert with over a decade of experience in the field.1 Here are some key details about Tianshi Gao:
Professional Experience
Current Role: As a Distinguished Engineer at Cruise, Tianshi is involved in developing autonomous vehicles using machine learning and artificial intelligence technologies.2 Cruise is focused on building advanced, all-electric, self-driving cars to safely connect people to places and experiences they care about.2
Previous Positions: Before joining Cruise, Tianshi held significant roles at Facebook:
- Senior Engineering & Research Scientist Manager (2016-2019)
- Senior Staff Research Scientist (2012-2016)
At Facebook, he led a core machine learning team of over 50 engineers, scientists, and managers working on ads ranking and personalization.2 His team's efforts contributed significantly to Facebook's ad revenue growth.
Education
Tianshi Gao has an impressive educational background:
- Ph.D. in Electrical Engineering from Stanford University
- Master's degree from Stanford University
- Bachelor's degree in Electrical Engineering from Tsinghua University (graduated with honors)1
Research and Publications
Tianshi is a prolific researcher and author:
- Published nine papers in five years
- Three papers selected for presentation at NeurIPS and ICCV
- Published as first author in top computer vision and machine learning conferences, including CVPR, ICCV, ECCV, ICML, and NeurIPS1
Achievements
- Contributed to significant improvements in Facebook's ad performance and revenue growth
- Developed fundamental machine learning and AI technologies in various areas, including sparse neural network architectures and distributed training algorithms1
- Holds patents based on his work during an internship at Dolby Laboratories2
Tianshi Gao's expertise in machine learning and AI, combined with his experience in leading large teams and driving technological advancements, makes him a valuable asset in the field of autonomous vehicle development at Cruise.
Highlights
The year 2021 has now come to an end. Looking back at it, my most memorable moment is a magical experience of the future. On Nov. 11th, Disco, a Cruise self-driving robot car, drove me on the public street of San Francisco, while I was busy taking a video of this silicon-based and invisible driver in action. I’ve probably spent over half a million minutes in a car, but this quarter of an hour just feels categorically and fundamentally different. It was surreal.
The magic of it is not only that a car drives itself but also how well it does it. Having been so used to human’s imprecise driving, I was pleasantly surprised by the gentleness, smoothness and consistency of every acceleration and deceleration at every intersection in this driverless ride. As someone with motion sickness, I feel that I can finally relax and be productive in a car.
But there is no magic. Just one week before my first driverless ride, we shared some of the under the hood innovations and hard works that make this possible. The full event can be watched here https://lnkd.in/gsSHYvgn. Below are just a few of my favorite details.
1/ Cruise AV (Autonomous Vehicle) is designed not to just replicate humans but to surpass them. This means that our AV is not artificially constrained to only optical sensors like humans but is augmented with additional sensing modalities. This allows our AV to see when optical sensors fail, e.g. inclement weather, dark night, leading to superhuman safety.
2/ Our AV does not only mechanically see and detect objects around it, but also anticipates the unseen based on the semantic understanding of the scene. It reasons about occlusions, e.g. a person pushing a garbage bin may suddenly appear behind a double parked garbage truck, thus drives cautiously proportionally to the unseen risks.
3/ Our AV also understands and anticipates possibilities beyond rules of road. It predicts bikes blowing through an all way stop intersection and cars cutting in wanting to turn from a straight-only lane. In general, our AV accounts for all sorts of behavioral uncertainties of other road users due to the partially observable nature of the world, be it the route uncertainty, control uncertainty or interaction uncertainty.
4/ From a machine learning standpoint, there are innovations in all aspects.
(I'm exceeding the max limit of characters. Please continue to see the details of how we are building the CNN and AlphaGo of self driving in the comments below.)
