Neural network behind Google Translate is (maybe) learning a type of own language

image01

The neural network behind the Google Translate service is under suspect to create a common representation of meaning, independend from language. Even for the Google engineers it seems to be hard to understand what is exactaly going on, but it is working with a measurable higher quality than ever before. The network was able to translate from Spanish to Portuguese by itself, without guidance, after it was trained to translate from Portuguese to English and English to Spanish.

Deep neural networks and machine-learning are key players of artificial intelligence. They are simulating basic information processing of the brain and are more and more used in many products.

More information: research.googleblog.com

Text summarization with the help of deep learning

The google Brain team open-sourced the TensorFlow model code for generating news headlines. Below are some examples. The first column shows the model input, which is the first sentence of a news article. The second column shows the headline generated by the algorithm.

dnn

Deep neural networks and machine-learning are key players of artificial intelligence. They are simulating basic information processing of the brain and are more and more used in many products.

More information: research.googleblog.com

A.I. foosball-table

Some students (computer engineering) at the Brigham Young University created an artificial intelligence foosball-table which is able to beat human foosball-players. The software tracks with the help of a camera (positioned over the table) the ball and operates the rods and players – and behaves like a human player: It anticipates, kicks and even scores.

This project impressively shows that computers do have the chance to assume human tasks – and sometimes operate much more faster.

More information: https://news.byu.edu