Synced Global AI Weekly | 2018.7.28—8.3

These Self-Driving Companies Never Stop Competing

Waymo Begins Experimenting with Self-Driving Taxi Prices

Waymo, the self-driving unit of Google parent Alphabet, has kept mum about how much it will eventually charge people to ride in its autonomous taxis. 

For more details

Tesla Is Building Its Own AI Chips For Self-Driving Cars

“We’ve been in semi-stealth mode on this basically for the last 2-3 years,” said Elon Musk on an earnings call today. “I think it’s probably time to let the cat out of the bag…”

For more details

Here We Go, Frisco; Our Self-Driving Service is Live, With One Million Simulated Miles Under Our Belt

"We announced the news of our self-driving service back in May, and now the time has arrived: today the pilot is launching in Frisco, Texas."

For more details

Uber to Stop Developing Self-Driving Trucks

Uber will stop developing self-driving trucks that have been hauling cargo on U.S. highways, the ride-hailing company said on Monday, seeking to focus its autonomous-vehicle technology solely on cars.

For more details


Fairness Matters: Promoting Pride and Respect with AI

"We think everyone should be able to express themselves online, so we want to make conversations more inclusive. That’s why we created tools like Perspective, an API that uses machine learning to detect abuse and harassment online."

For more details

GANimation: Anatomically-aware Facial Animation from a Single Image

"In this paper, we introduce a novel GAN conditioning scheme based on Action Units (AU) annotations, which describes in a continuous manifold the anatomical facial movements defining a human expression."

(arXiv) / (Github)

Learning Dexterity

"We've trained a human-like robot hand to manipulate physical objects with unprecedented dexterity. Our system, called Dactyl, is trained entirely in simulation and transfers its knowledge to reality, adapting to real-world physics using techniques we’ve been working on for the past year."


You May Also Like

Synchronized SGD Outdated? Yoshua Bengio on Hardware-Friendly DL

DeepMind’s 2018 AlphaGo Zero requires 300,000 times more computing power than AlexNet did in 2013. With larger-than-ever datasets and demanding high precision models to deal with, Deep Learning (DL) algorithms are understandably hungry for new hardware solutions...

For more details

One Billion Billion! Tianhe-3 Exascale Supercomputer Prototype Passes Tests

Scientists at the National Supercomputing Center of Tianjin in China have unveiled their prototype of a next-generation exascale supercomputer. The centre says the Tianhe-3 can perform at one exaFLOPS, or a billion billion calculations per second - the threshold for coveted “exascale” status...


Global AI Events

5–10 AugTDWI Anaheim Conference.Anaheim, USA
6–10 AugUAI 2018Monterey, USA
19–23 AugKDD 2018London, UK
12–16 AugSIGGRAPH 2018Vancouver, Canada
20–24 AugCASEMunich, Germany
20–24 AugICPRBeijing, China

Global AI Opportunities

Aquifi needs Machine Learning Scientist, Computer Vision Engineer and Mechanical Design Engineer in Palo Alto, CA.

Apply Now

GoogleAI in Ghana is recruiting Machine Learning Research Scientist, Research Software Engineer and AI researcher.

Apply Now

Appthority is hiring Senior Software Engineer for Data System, Front-end, Cloud Back-end, Android and iOS in Mountain View, CA.

Apply Now

Endgame wants Software Engineer (Frontend/Backend) and Site Reliability Engineer in San Francisco, CA and Arlington, VA.

Apply Now


从 20 世纪 80 年代首次成功演示以来(Dickmanns & Mysliwetz (1992); Dickmanns & Graefe (1988); Thorpe et al. (1988)),自动驾驶汽车领域已经取得了巨大进展。尽管有了这些进展,但在任意复杂环境中实现完全自动驾驶导航仍被认为还需要数十年的发展。原因有两个:首先,在复杂的动态环境中运行的自动驾驶系统需要人工智能归纳不可预测的情境,从而进行实时推论。第二,信息性决策需要准确的感知,目前大部分已有的计算机视觉系统有一定的错误率,这是自动驾驶导航所无法接受的。




阿尔法围棋是于2014年开始由英国伦敦Google DeepMind公司开发的人工智能围棋程序。AlphaGo是第一个打败人类职业棋手的计算机程序,也是第一个打败围棋世界冠军的计算机程序,可以说是历史上最强的棋手。 技术上来说,AlphaGo的算法结合了机器学习(machine learning)和树搜索(tree search)技术,并使用了大量的人类、电脑的对弈来进行训练。AlphaGo使用蒙特卡洛树搜索(MCTS:Monte-Carlo Tree Search),以价值网络(value network)和策略网络(policy network)为指导,其中价值网络用于预测游戏的胜利者,策略网络用于选择下一步行动。价值网络和策略网络都是使用深度神经网络技术实现的,神经网络的输入是经过预处理的围棋面板的描述(description of Go board)。




AlexNet是一个卷积神经网络的名字,最初是与CUDA一起使用GPU支持运行的,AlexNet是2012年ImageNet竞赛冠军获得者Alex Krizhevsky设计的。该网络达错误率大大减小了15.3%,比亚军高出10.8个百分点。AlexNet是由SuperVision组设计的,由Alex Krizhevsky, Geoffrey Hinton和Ilya Sutskever组成。