MIET master participated in the international Conference on Neural Information Processing Systems in Canada

From 7 to 12 of December Montreal (Canada) hosted the Twenty-ninth Annual Conference on Neural Information Processing Systems (NIPS). It is a single-track machine learning and computational neuroscience conference that includes invited talks, demonstrations and oral and poster presentations of refereed papers. The conference was attended by more than 3 thousand people.

Master of the department of Higher Mathematics Alexander Fedorov presented the work Deep Attention Recurrent Q-Network at the forum Deep Reinforcement Learning Workshop. The work was dedicated to the use of neural network mechanisms of Hard and Soft Attention as an extension of the architecture of the neural network of deep Deep Q-Network.

Deep Q-Network is a universal neural network algorithm of learning, with the addition published in early 2015 in the prestigious scientific journal “Nature”. Deep Q-Network architecture, developed by researchers at Google DeepMind, learned to play classic games Atari 2600. In most cases, trained artificial intelligence surpasses human capabilities.

See how the computer plays Atari Games.