Home / Technology / Google’s new artificial neural network “PlaNet” outperforms everything in locating places

Google’s new artificial neural network “PlaNet” outperforms everything in locating places

Search engine giant Google has designed a new artificial intelligence system that is exceeding every other system in spotting the source of images. This new system is extremely good at locating identifying the location where photos were taken .Google named the new system as “PlaNet”. The system is known as a deep learning neural network, the more images it receives, the more intelligent it gets.

The research team that worked hard behind the implementation of “PlaNet” commented that PlaNet is such a smart tool that it is able to localize 3.6% of the images with street level accuracy, 10.1% at city level accuracy, 28.4% at country level and 48.0% at continent level.

The research team organized a competition to identify the effectuality of the software. They compared PlaNet to human accuracy. 10 well traveled people were invited to have a play off session against PlaNet. PlaNet singlehandedly won 28 rounds out of 50 rounds of the game. In the game” Geoguessr” PlaNet scored a median localization error of 1131.7 km and the median human localization error was 2320.75 km.

The research paper reveals that “PlaNet” is handling images with high accuracy by simply following a method. For example, if “PlaNet” was trying to identify a forest location, then it would match millions of photos of the forest and the one looks most identical would be regarded as the location of the forest. In this way, with sufficient data on reach, this mechanism can reach high accuracy.   Google engineers have created a gigantic database containing 126 million geolocated photographs amalgamated from the internet. These images were matched with the place those were taken at. Google have managed to train their super computer to indentify the place where the images were taken but looking at its pixel.

The team of researchers arranged 29.7 millions of public photos to check out the efficiency of the neural network. Google collected these photos from Google. However Google also revealed few limitations of the program. Because the programs performance solely depends on the collection of Google images, the program suffers when it is confronted with rural countryside photographs. Because the program requires much less space to be installed in smart phones, it can be installed with 377 MB .Because of the efficiency level, “PlaNet” is considered as the Superhuman.

About Nandita Das

Leave a Reply

Your email address will not be published. Required fields are marked *