This announcement should alarm a little more all those who fear progress these last months in artificial intelligence. The consequences of the work of researchers from the University of Maryland and the NICTA Australian research centre are potentially enormous. In a few days, they will present the results of their research at the next conference of the AAAI (Association for the Advancement of Artificial Intelligence). These researchers in artificial intelligence are able to teach a robot Baxter to cook… just watching videos of kitchen on YouTube!
An application of Machine Learning: decrypt YouTube videos
Today to learn how to use software, Discover tips and tricks for tinkering or learn a recipe, the best way is still to take a look on YouTube to find a tutorial. And that is perhaps what will someday machines to imitate us. Researchers have thus analyze scenes YouTube to a robot to teach him to reproduce recipes.
Their research is based on the CNN analysis (” convolutional neural network”»»), a Machine Learning technique well known for the recognition of forms. This allows the robot to identify objects and movements in the video arena. This image analysis, researchers have added algorithms of elements of the video scene as well as a semantic algorithm in order to connect these actions between them and allow the robot to replicate what he saw on the screen.
Researchers will present the results of their work at the end of this month. If this approach keeps its promises, they will have made a significant step in what remains a major problem of modern robotics, their learning.
Translation : Bing Translator
“Robots can now learn to cook just like you do: by watching YouTube videos”, VentureBeat, January 3, 2015