We use deep neural networks to assess the states of a relevant environment. Environments may range from positions on a chessboard to configurations of manufacturing processes to disposition of airplanes over international airports worldwide. Once the neural network has learned how to assess the states, one can use it to make optimal decisions that maximize one’s reward in a long run (win a game, minimize manufacturing costs, optimize flight schedules, etc.).

Principle Investigators: Pavel Gurevich, Hannes Stuke
Members: Florian Dorner, Dmitry Puzyrev (Associate member)