Deep-learning computers are advancing toward true artificial intelligence (AI) that will enable them to think as humans do. The approach relies on tremendous data sets and vast computing power to answer problems that humans can easily solve, such as identifying patterns in a large number of images to identify categories such as cats and people. Deep learning is based on the concept of neural networks, which are modeled loosely on the interconnected neurons of the human brain. Progress in deep learning is bringing consumers software that is better able to sort through photos, understand spoken commands, and translate text from foreign languages. Furthermore, scientists are using deep-learning computers to identify potential drug candidates, map neural networks in the brain, and predict the functions of proteins. Deep learning in neural networks appeared promising in the 1980s, but enthusiasm dwindled when the approach proved challenging. However, interest resumed in 2000 due to increases in computing power and a sudden abundance of digital data, with many researchers focusing on speech and image recognition. Now the emphasis in deep learning is shifting to natural language understanding to enable computers, for example, to comprehend human speech well enough to rephrase and answer questions, or to translate languages. Although the results show great potential, deep learning is still a nascent field.
More info: Nature (01/08/14) Nicola Jones