Computer Vision News - April 2022

27 AI Spotlight News Building a Hybrid AI Able to Beat the World's Best Bridge Player For the son of two terrific bridge players (and a very poor player myself), this is quite special: a French private AI lab called Nukkai has been working on an AI that may just be able to beat the world's best bridge player , something that has already been made possible in other games like Chess and Go. This is no simple matter: in chess and go, competitors play with complete information and must react to the behavior of a single opponent at a time, while everyone is in possession of all the information. In bridge, the opposite is true: multiple opponents and only partial information - a scenario much closer to human decision-making. Watch the Video insideBIGDATA: The Decade of Synthetic Data is Underway The folks at insideBIGDATA publish a nice tribune (signed by Datagen’s CTO), where it is declared that the 2020s will be remembered as the “Decade of Data” for AI, and - even more - the “Decade of Synthetic Data” . As we have ourselves explained a short time ago , deep learning-based systems require sufficient data for proper training and reliable testing. A solution is to generate data synthetically : the nice thing is that it works, and it has gained widespread acceptance across the research and enterprise communities. Probably because it is much simpler/faster than spending hundreds of hours on fine-tuning AI algorithms and models. Read More Watch a Robot Peel a Banana Without Crushing It into Oblivion Handling eggs and soft fruit is tricky for robots, but apparently, after training with hours of data from a human operator, a machine learning algorithm developed the motor skills needed to correctly peel a banana. Not always, but most of the times and in less than 3 minutes. The effort was done by Heecheol Kim at the University of Tokyo : the machine-learning system that he and his colleagues have developed powers a robot, which has two arms and hands that grasp between two “fingers”. 13 hours of deep-imitation learning were enough to train the robot. Watch the Video

RkJQdWJsaXNoZXIy NTc3NzU=