A team of researchers from the laboratory Deep Mind Google has developed an intelligent algorithm capable of learning to play by himself!
Called DQN it a little special software was developed by a team of researchers from the laboratory Deep Mind of Google, specializing in artificial intelligence. The objective was to create an algorithm capable of learning to play by himself. To do this, DQN has been tested on 49 games of the Atari 2600 and would be able to beat the record of professional players in 23 of them.
And where others existing systems merely recite predefined commands or enforce rules already known, this algorithm evolves to a whole other level. Indeed, before starting a game, it does not know the rules. It is as parties and analyzing the operation of the game that improves and develops strategies in the manner of a human being. It thus eventually reach an impressive level at Breakout particular, since it includes that caught the ball at the top of the screen allows it to break more bricks in a row and thus achieve a better score faster.
Obviously, the system has its limitations and DQN was more effective on "simple" games. Indeed, to understand more elaborate games and using more complex mechanical such as Ms Pac-Man, Asteroids or Montezuma's Revenge was a lot more complicated for him and realized that there is poor scores or no. The algorithm is not suitable for complex and recent games and is not yet able to analyze rich 3D environments, or to determine the sequences of complex games.
However, the team behind the project believes that his algorithm will be able eventually to find more sophisticated ways and have diverse applications, such as medical imaging or automated lead.
I founded the Blog DepoTech in January 2015. Self-taught in development, I have always pushed my curiosity about the topics,Web and Tech news.
ConversionConversion EmoticonEmoticon