Deep neural nets (DNNs) contain multiple intermediate layers of artificial neuron units (e.g., ranging from just a few layers to dozens of layers), which increase the ability of DNNs to learn. DNNs comprise the “architecture” upon which deep learning takes place.

1965

DEEP LEARNING

Artificial intelligence involves approaches that enable machines to mimic human intelligence. Machine learning (see page 99) is a form of AI that enables machines to improve at some task through practice and experience. Deep learning is a form of machine learning that allows systems to train themselves to perform tasks, like playing a game or recognizing a cat in a photo, using deep neural nets (DNNs), which have multiple intermediate layers of artificial neuron units, in contrast to shallow nets, which employ just a few layers. Although the phrase deep learning was not introduced until 1986, Soviet mathematician Alexey Ivakhnenko (1913–2007) conducted early work, in the form of supervised deep multilayer perceptrons, in 1965.

Generally speaking, the multiple layers of neurons can perform feature extraction on the data at different levels of a hierarchy (e.g., responding to simple edges at one level and to facial features at another). Deep learning may involve backpropagation, a process in which the system can pass information in a reverse direction, from output to input, in order to teach the system when it has made errors so as to improve the results.

Deep learning has been successfully applied to speech recognition, computer vision, natural language processing, social networking, human language translations, drug design, identifying specific style periods for paintings, making product recommendations, estimating the value of different marketing actions, performing image restoration and cleaning, playing games, identifying people in photos, and more.

As technologist Jeremy Fain writes: “Ultimately, deep learning has pushed machine learning across a threshold. Whereas machine learning had some success in automating repetitive tasks or data analytics, it is now bringing the future to life in the form of computers that can see, hear and play all types of games.”