Deep learning is a subset of machine learning, which is itself a type of artificial intelligence, or AI. In deep learning, computers use artificial neural networks that are designed with the rough structure of a human brain, allowing the machines to process incoming data in a non-linear fashion. As new data comes into the neural network, the computer incorporates it, finds patterns, and improves its own processing system. Because they use a web-like structure that allows non-linear “thinking,” deep learning systems can handle a greater amount of input than traditional linear data analysis.
Deep learning software is designed to mimic the way the brain thinks. This begins with learning to recognize patterns, whether they come via images, sounds or other data. To accomplish this, enormous artificial neural networks are needed to create that “web” that allows learning. The systems are called “deep” because they’re designed in layers, with processing of incoming data occurring simultaneously on multiple layers at the same time, just as a human brain works. Once these networks are established, massive amounts of data can be employed to begin training the system to recognize patterns, and learning begins.
Google has been at the forefront of the quest to expand deep learning. Its recent triumph involved teaching a computer to categorize images from YouTube videos correctly in essence, training the computer to recognize a cat when it saw one. This effort involved 16,000 computer processors and 10 million YouTube video clips, and the results were twice as good as any previous attempt. Deep learning is also at the heart of work being done to improve speech recognition and machine translation. Researchers predict that the future of deep learning will take it into the medical, robotics and linguistic fields as well.