2013 Breakthrough Science and Technology "deep learning"

In July 2012, Rey Kuzwer with Google CEO Larry Page when he wasn't looking for a job. Kurzweil, the famous mechanical intelligence, told Paige that he had studied the whole computer system, and that he had a preliminary plan to start a company to build such computers. He refers to a computer that is capable of understanding the language himself, making inferences and making his own decisions.

To complete this work, it is natural to use Google's data and computing power. "I can get you data and servers, but it's too hard for a company to do it," Page told him, adding that Kurzweil, who had never worked for someone else, had joined Google. After 6 months of struggle, Kurzweil finally chose to be the project director's identity to join Google. "This is the culmination of more than 50 years of research on AI," he said. ”

What attracts Kurzweil to Google is not just Google's unique data and powerful supercomputers, but the "deep learning" of a gleaming AI branch within Google Inc. Google's deep learning software tries to mimic the neuronal activity in the human cortex, the brain of 80% in the region's head. This software can identify digitized sounds, pictures, and other data fragments.

The basic idea of neural network has appeared for decades, but it has not made much breakthrough. However, with improved algorithms and computer performance, scientists can now simulate more nerve-layer neural activity.

With deep learning, they have made rapid progress in the field of speech recognition and image recognition. In the June 2012, Google's in-depth learning system improved the accuracy of objects by one-fold, and drastically reduced the error rate of the Android system's speech recognition system. In December 2012, Microsoft demonstrated a Chinese-English instant interpretation system with an error rate of only 7% and a smooth pronunciation. Also in December, a group of college students and two professors made use of deep learning software to accomplish molecular recognition, which could be used to discover new medicines for treatment.

Google is now becoming a "deep learning" of the magnet, attracting high-end professionals from around the world. March 2013, Google acquired a deep learning enterprise. Founded by Jeffrey Hitton, a computer science professor at the University of Toronto, Geoffrey Hinton, the company was awarded the Merck Prize. Heaton now divides his time in half, giving it to the university and half to Google. His plan was to "take the theory of the university and apply it to practical issues", such as image recognition, search, natural language understanding, and other related artificial intelligence.

The area of application mentioned above is a reminder to AI researchers that smart machines in sci-fi movies are very promising in real life. Indeed, the application of machine intelligence is shifting from communication and computing to medical, manufacturing and transportation. For example, IBM's deep learning technology is being used to train physicians to help them make the right choices; Microsoft's deep learning technology is used in Windows Phone and Bing voice searches.

However, expanding the depth of learning applications from image and speech recognition to other areas requires greater conceptual and software breakthroughs, as well as a further increase in computing power for computers. Maybe within a few years we will not see the full computer, but there are no problems with such computers in a few decades. "Deep learning has stirred up new challenges in the field of artificial intelligence," said Peter Lee, dean of Microsoft's American Institute. ”

(Responsible editor: Fumingli)