Deep neural networks

A major advance in AI since 2010 has been the invention of Deep Neural Networks (DNN). By 2006, acoustic models for speech recognition had been significantly improved thanks to these networks. Since then, they have demonstrated their effectiveness in other fields: in games, in natural language processing, in computer vision, in diagnostics, in robotics, in art, and so on.

Deep networks owe their name to the large number of hidden layers of neurons, which can number in the hundreds. This gives them the ability to learn ever more complex matching functions. There are different types of deep networks: feedforward perceptrons, recurrent networks, convolutional networks. For a detailed description, please see